{ "version": "1.0", "verified_at": "2026-10-14", "source": { "id": "models-dev", "name": "models.dev", "repo": "anomalyco/models.dev", "commit": "59aa52dbc09727fa6ead433b0285970ebc05ba82", "url": "https://github.com/anomalyco/models.dev", "license": "MIT" }, "models": [ { "id": "model-aisingapore-gemma-sea-lion-v4-27b-it", "source_id": "aisingapore/gemma-sea-lion-v4-27b-it", "name": "Gemma-SEA-LION-v4-27B-IT", "developer": "AI Singapore", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT", "description": "AI Singapore's Gemma-SEA-LION-v4-27B-IT, a Gemma-3-27B-based multimodal instruct model post-trained for Southeast Asian languages with RLHF and 128K context, published as ungated open weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Gemma-SEA-LION-v4-27B-IT", "description": "Gemma 3 27B tuned by AI Singapore for Southeast Asian languages and instruction following", "family": "gemma", "release_date": "2025-09-23", "last_updated": "2025-09-23", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 128000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT" } ] }, "licenses": [ "LicenseRef-Gemma-Terms-of-Use", "MIT" ], "source_model": "mixed_source", "license_review_status": "verified", "license_note": "The weights are governed by the Gemma terms of use because the release is built on Gemma 3; the repository-root MIT license file covers the repository code and software only. Both material licenses are recorded. The release is explicitly not safety-aligned and the v4.5 line supersedes this v4 release.", "license_evidence": [ { "license_id": "LicenseRef-Gemma-Terms-of-Use", "scope": "Gemma-SEA-LION-v4-27B-IT weights built on the Gemma 3-27B base under Gemma terms of use", "kind": "web_terms", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT", "verified_at": "2026-09-22" }, { "license_id": "MIT", "scope": "Repository-root LICENSE file covering the repository code and software, not the model weights", "kind": "web_terms", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Gemma-SEA-LION-v4-27B-IT weight release post-trained in 2025 for Southeast Asian languages, a Gemma-3-27B-based instruct model. The Gemma base models, the base SEA-LION checkpoint, the v4.5 successor line, and independent hosts are separate boundaries.", "strengths": [ "Ungated 128K-context multimodal release with SEA-language post-training and RLHF documented for eight regional languages", "Weight links work and serving is available through an inference provider and HuggingChat" ], "tradeoffs": [ "Explicitly not safety-aligned, and vision capability is inherited from Gemma 3 rather than post-trained", "Dual-layer licensing (Gemma terms for weights, MIT file for code) needs careful evidence scope" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.32 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/aisingapore/gemma-sea-lion-v4-27b-it.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-ai2-olmo-3-1-32b-think", "source_id": null, "name": "Olmo 3.1 32B Think", "developer": "Ai2", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "description": "Ai2's December 2025 Olmo 3.1 reasoning flagship at 32B parameters with explicit reasoning chains, published fully open under Apache 2.0 with weights, training data, code, and intermediate checkpoints on Ai2's Hugging Face organization alongside a technical report. No developer API serves this checkpoint; the playground is a demo. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Olmo 3.1 32B Think", "description": "Ai2's most performant 32B reasoning checkpoint, extending the Olmo 3 Think reinforcement-learning run with a longer training schedule", "family": "olmo", "release_date": "2025-12-10", "last_updated": "2025-12-10", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": null, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The allenai/Olmo-3.1-32B-Think model card declares license apache-2.0 in its frontmatter and states the model is licensed under Apache 2.0, intended for research and educational use in accordance with Ai2's Responsible Use Guidelines. The repository ships no LICENSE file of its own, so the grant is the standard Apache 2.0 text plus that accompanying use guidance.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Olmo-3.1-32B-Think weights published in Ai2's Hugging Face repository under the Apache 2.0 licence declared by the model card", "kind": "web_terms", "url": "https://opensource.org/licenses/Apache-2.0", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the Olmo 3 line's 32B reasoning identity at its current Olmo 3.1 Think checkpoint. The Olmo 3.0 Think and Base checkpoints, the 7B variants, Olmo 3.1 Instruct 32B, the Tulu and Molmo lines, and the Ai2 playground demo are separate boundaries.", "strengths": [ "Fully open model flow: weights, Dolma training data, training code, and intermediate checkpoints all published under Apache 2.0", "Explicit reasoning chains in the flagship 32B class with a technical report, model card, and developer documentation from the institute" ], "tradeoffs": [ "The 32B checkpoint needs accelerator-class serving hardware, and no developer API serves it, so running it is entirely self-operated", "The card steers the model toward research and educational use under Ai2's Responsible Use Guidelines rather than unrestricted deployment" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 2, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.16 }, "evidence": [ { "kind": "web", "label": "Ai2 Olmo 3 announcement: November 20, 2025 family release with the December 3.1 update", "url": "https://allenai.org/blog/olmo3", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Ai2 Olmo-3.1-32B-Think weight repository: apache-2.0 licence, text generation", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Olmo 3 technical report", "url": "https://allenai.org/papers/olmo3", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-ai21-jamba-large", "source_id": "ai21/jamba-large", "name": "Jamba Large 1.6", "developer": "AI21 Labs", "url": "https://huggingface.co/ai21labs/AI21-Jamba-Large-1.6", "description": "AI21 Labs' Jamba Large 1.6, a hybrid SSM-Transformer enterprise language model with 256K context, published under the Jamba Open Model License on a gated Hugging Face repository.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Jamba Large", "description": "AI21's hybrid SSM-Transformer long-context model for enterprise agents and grounded generation", "family": "jamba", "release_date": "2025-07-01", "last_updated": "2025-07-01", "knowledge_cutoff": "2024-08-22", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/ai21labs/AI21-Jamba-Large-1.6" } ] }, "licenses": [ "LicenseRef-Jamba-Open-Model-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The Jamba Large 1.6 weights are governed by the Jamba Open Model License, documented on AI21's license page and in the repository LICENSE.txt. The custom license is not OSI-approved: it carries a $50M revenue gate, attribution and naming requirements, and revocability.", "license_evidence": [ { "license_id": "LicenseRef-Jamba-Open-Model-License", "scope": "AI21-Jamba-Large-1.6 repository weights and checkpoints under the Jamba Open Model License", "kind": "web_terms", "url": "https://www.ai21.com/licenses/jamba-open-model-license", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the AI21-Jamba-Large-1.6 checkpoint released March 6, 2025. The Jamba Mini 1.6 sibling, earlier Jamba checkpoints, and third-party hosts are separate boundaries.", "strengths": [ "The first Jamba record in the Atlas, a distinct hybrid SSM-Transformer architecture with 256K context", "The governing license is documented on an official page and in the repository license file" ], "tradeoffs": [ "The custom Jamba license is not open source, with a revenue gate and revocability", "The weight repository is gated and requires accepting terms" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 6, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 5.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/ai21/jamba-large.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/ai21labs/AI21-Jamba-Large-1.6", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Jamba 1.6 announcement", "url": "https://www.ai21.com/blog/introducing-jamba-1-6", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-ai21-jamba-mini", "source_id": "ai21/jamba-mini", "name": "Jamba Mini 1.6", "developer": "AI21 Labs", "url": "https://huggingface.co/ai21labs/AI21-Jamba-Mini-1.6", "description": "AI21 Labs' Jamba Mini 1.6, the 52-billion-parameter sibling checkpoint of the Jamba 1.6 line with 256K context, function calling, and structured output, published under the Jamba Open Model License.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Jamba Mini", "description": "AI21's efficient, lightweight hybrid SSM-Transformer model for a wide range of tasks", "family": "jamba", "release_date": "2026-01-01", "last_updated": "2026-01-01", "knowledge_cutoff": "2024-08-22", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/ai21labs/AI21-Jamba-Mini-1.6" } ] }, "licenses": [ "LicenseRef-Jamba-Open-Model-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The Jamba Mini 1.6 weights are governed by the same Jamba Open Model License as the Large sibling, documented on AI21's license page and in the repository LICENSE.txt.", "license_evidence": [ { "license_id": "LicenseRef-Jamba-Open-Model-License", "scope": "AI21-Jamba-Mini-1.6 repository weights and checkpoints under the Jamba Open Model License", "kind": "web_terms", "url": "https://www.ai21.com/licenses/jamba-open-model-license", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the AI21-Jamba-Mini-1.6 checkpoint. The Jamba Large 1.6 sibling, earlier Jamba checkpoints, and third-party hosts are separate boundaries.", "strengths": [ "A distinct sibling checkpoint of the Jamba 1.6 line with a separate size and record boundary", "256K context with documented function calling and structured output" ], "tradeoffs": [ "Same custom-license and gating constraints as the Large sibling", "Separate record from Large per the Atlas size-sibling convention, so readers must compare the pair" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 6, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 5.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/ai21/jamba-mini.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/ai21labs/AI21-Jamba-Mini-1.6", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "access_boundary": "This record represents the AikidoSec/altar-1 Altar-1 release announced September 21, 2026. The zai-org/GLM-5.3 parent release, the cyankiwi/GLM-5.3-AWQ-INT4 quantization base, 0xSero's sibling 569B and EXL3 cuts of the same expert selection, the Aikido Machine on-prem appliance, the Aikido Attack, Code Audit, and Deep Review products, and any hosted inference endpoint remain separate boundaries.", "description": "Aikido Security's first open-weight release: a 168-expert REAP prune of the Z.AI GLM-5.3 mixture-of-experts model at AWQ INT4 (328 GB) for self-operated defensive-security workloads, published under the inherited custom GLM-5.3 License carrying a large-operator model-as-a-service security-review condition, without a developer API.", "developer": "Aikido Security", "distribution_modes": [ "downloadable_weights" ], "evidence": [ { "kind": "web", "label": "Official Aikido Altar announcement: method, storage, evaluation, and serving", "url": "https://www.aikido.dev/blog/aikido-altar-open-weight-ai-sovereign-security", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Developer-published AikidoSec/altar-1 weights repository and model card", "url": "https://huggingface.co/AikidoSec/altar-1", "verified_at": "2026-09-22" }, { "kind": "web", "label": "GLM-5.3 License file inherited by Altar-1, with model-as-a-service security-review clause", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "id": "model-aikido-altar-1", "license_evidence": [ { "kind": "web_terms", "license_id": "LicenseRef-GLM-5.3", "scope": "AikidoSec/altar-1 weights, configuration, tokenizer, and documentation by inheritance of the GLM-5.3 License; the Altar-1 repository ships no separate license file. Hosted Aikido products and services are governed separately.", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "license_note": "The AikidoSec/altar-1 model card states the release inherits the GLM-5.3 license rather than shipping its own license file. The governing text is the zai-org/GLM-5.3 license granting use, copying, modification, distribution, deployment, and fine-tuning, but requiring any licensee operating a model-as-a-service business above 10 billion US dollars in aggregate affiliate revenue over any consecutive twelve months to pass a Z.AI security review before any commercial use. That gate breaks from an open-source posture, so the release is source-available rather than open source.", "license_review_status": "verified", "licenses": [ "LicenseRef-GLM-5.3" ], "metadata_verified_at": "2026-09-22", "model_type": "language_model", "name": "Altar-1", "score": { "artifact_availability": 8, "deployment_portability": 6, "documentation_provenance": 8, "license_clarity": 5, "lifecycle_transparency": 6, "overall": 6.14, "serving_reach": 4 }, "score_profile": "model_access", "source_id": null, "source_metadata": { "capabilities": { "attachment": null, "reasoning": true, "structured_output": true, "temperature": null, "tool_call": true }, "description": "Compressed open-weight AI model for sovereign security intelligence, powering on-prem air-gapped pentesting", "family": "altar", "knowledge_cutoff": null, "last_updated": null, "limits": { "context": 131072, "input": null, "output": null }, "links": [], "modalities": { "input": [ "text" ], "output": [ "text" ] }, "name": "Altar-1", "release_date": "2026-09-21", "reported_license": "other", "reported_open_weights": true, "weights": [] }, "source_model": "source_available", "strengths": [ "Developer-published sharded W4A16 safetensors checkpoint with tokenizer, chat template, and prune plan on the AikidoSec Hugging Face organization, plus a documented vLLM command for a 4xH200 node", "First-party announcement and model card document the REAP pruning method, the per-checkpoint storage table, the calibration domains, and an internal 32-vulnerability rediscovery comparison with its pipeline boundaries stated" ], "tradeoffs": [ "The inherited custom GLM-5.3 License gates commercial model-as-a-service use by large operators behind a Z.AI security review", "The documented self-hosted path requires Hopper GPUs in a 4xH200 configuration; no developer API or third-party hosted path is documented", "No versioning or deprecation posture is stated for the Altar line" ], "url": "https://huggingface.co/AikidoSec/altar-1", "verified_at": "2026-09-22" }, { "id": "model-aleph-alpha-kolibri-1", "source_id": null, "name": "Kolibri 1", "developer": "Aleph Alpha Research GmbH", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1", "description": "Aleph Alpha's sovereign English-German mixture-of-experts reasoning model, 78.1B total parameters with 3.46B active per token, trained from scratch for public administration and regulated industry and published as FP8 weights under Apache 2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Kolibri 1", "description": "Kolibri is Aleph Alpha's mixture-of-experts (MoE) reasoning model, with a focus on German and English. The model supports an explicit reasoning mode and tool calling. It is optimized for long-context and inference efficiency.", "family": "kolibri", "release_date": "2026-10-03", "last_updated": null, "knowledge_cutoff": "2026-06-18", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The Kolibri-1 model card states that \"The model weights are published by Aleph Alpha GmbH under Apache 2.0 license. The rights granted thereunder only apply to the weights and configuration files published in this repository. For the avoidance of doubt, any other artifacts not included in the repository are excluded from the license. The license especially does not extend to underlying code, model architecture, parameter settings or any training method.\" The repository ships the unmodified Apache License 2.0 text with no filled-in copyright line, and the card adds a General-Purpose AI Code of Practice point of contact for rightsholders. The card's Responsible Use section is guidance rather than a licence term: it opens by \"Acknowledging the permissive nature of the license under which the weights of Kolibri are released\" and asks users to refrain from unlawful use, so it adds no mandatory restriction to the artefacts the licence covers. The classification is of the published weights and configuration files, not of training code, data, or methods, which Aleph Alpha retains.", "license_evidence": [ { "kind": "web_terms", "license_id": "Apache-2.0", "scope": "The weights and configuration files published in the Aleph-Alpha/Kolibri-1 repository; the model card states the licence does not extend to code, model architecture, parameter settings, or training methods, which Aleph Alpha retains", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1/blob/main/LICENSE", "verified_at": "2026-10-03" } ], "access_boundary": "This record represents the Kolibri 1 release of 3 October 2026, the version whose weights are published as Aleph-Alpha/Kolibri-1 and whose serving parsers are named kolibri1. It excludes Kolibri Origin, the earlier pipeline-validating iteration the developer states had \"no public release\"; the Pharia enterprise platform and any managed endpoint, since no first-party API is documented for this release; the aleph-alpha-inference serving package, which is a separate repository; and any downstream fine-tune, quantization, or re-upload of these weights.", "strengths": [ "Developer-published FP8 safetensors weights under Apache 2.0 with the licence scope stated in the model card itself, covering the repository's weights and configuration files and expressly excluding code, architecture, and training methods", "A 189-page technical report and a model card that document the full pipeline, including 768 B200 GPUs over 21 days of pre-training, the bilingual data mix, more than 1.2M reinforcement-learning tasks, energy consumption, and per-checkpoint evaluation across safety, grounding, and abstention", "A documented first-party serving path: an aleph-alpha-inference package with a ghcr.io container image, a vLLM serve command using the Kolibri reasoning and tool-call parsers, recommended sampling parameters, and an OpenAI-compatible local endpoint" ], "tradeoffs": [ "The full 78B model must be resident even though 3.46B parameters activate per token, so the documented minimum is 2x A100 80 GB, 2x H100 SXM5, or one H200, B200, or B300", "The developer recommends contexts of at most 262,144 tokens for serving efficiency and complex tasks even though 1,048,576 tokens has been validated, and longer contexts need explicit vLLM overrides", "No first-party managed API is documented for this release; access beyond the weights is enterprise deployment through sales contact, and no third-party hosting path is documented", "Implicit knowledge stops at 18 June 2026 for both German and English, so deployments without tool access work from a dated world model" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.64 }, "evidence": [ { "kind": "web", "label": "Developer-published Kolibri-1 weights repository and model card: provider and developer entities, precision, context, knowledge cutoff, intended use, limitations, and the License and terms section", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1", "verified_at": "2026-10-03" }, { "kind": "web", "label": "Apache License 2.0 text published in the Kolibri-1 repository, governing the weights and configuration files", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1/blob/main/LICENSE", "verified_at": "2026-10-03" }, { "kind": "web", "label": "Official Kolibri model page: generally available status, mixture-of-experts architecture, 78B total and 3B active parameters, 1,048,576-token context, hardware requirements, Apache 2.0 licence line, and the open-weights and enterprise-deployment access routes", "url": "https://aleph-alpha.com/en/kolibri/", "verified_at": "2026-10-03" }, { "kind": "web", "label": "Release announcement of 3 October 2026: the Kolibri and Kolibri Origin comparison table, the Apache 2.0 terms, the 20T-token training progression, and the documented vLLM serving command", "url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/", "verified_at": "2026-10-03" }, { "kind": "web", "label": "Kolibri technical report: an English-German mixture-of-experts transformer with 78.1B total and 3.46B active parameters per token, released as open weights under the Apache 2.0 licence", "url": "https://aleph-alpha.com/downloads/tech-report.pdf", "verified_at": "2026-10-03" } ], "metadata_verified_at": "2026-10-03", "verified_at": "2026-10-03" }, { "id": "model-alibaba-qwen-plus-2025-12-01", "source_id": null, "name": "Qwen Plus 2025-12-01", "developer": "Alibaba", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen-plus", "description": "Alibaba's December 1, 2025 snapshot of Qwen-Plus, the Qwen3-series Plus text model with switchable thinking and non-thinking modes and 1M context, served through Alibaba Cloud Model Studio with no downloadable weights.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Qwen Plus 2025-12-01", "description": "Qwen-Plus snapshot with improved reasoning, agent, and multi-turn tool-calling capabilities over the July 28, 2025 snapshot", "family": "qwen", "release_date": "2025-12-01", "last_updated": "2025-12-01", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 1000000, "input": 997952, "output": 32768 }, "reported_open_weights": null, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Alibaba-Model-Studio-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The qwen-plus-2025-12-01 snapshot is served as a Model Studio API model under the Alibaba Cloud Model Studio Service Agreement, effective September 2, 2026, which Model Studio's related-agreements page lists as governing the service. It is a paid-service contract that bars reselling the service or its models and using them or their outputs to train or develop competing products without written permission, and prohibits reverse engineering or trying to discover the service's source code. It grants no licence to model weights for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Alibaba-Model-Studio-Agreement", "scope": "Business and developer access to the qwen-plus-2025-12-01 snapshot through Alibaba Cloud Model Studio API services", "kind": "web_terms", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" } ], "access_boundary": "This record represents the Qwen-Plus line at its current pinned snapshot, qwen-plus-2025-12-01, which Alibaba Cloud Model Studio serves with 1M context and to which the undated qwen-plus model resolves. The earlier dated Qwen-Plus snapshots, the dynamically updated qwen-plus-latest alias, the Qwen3.5, 3.6, and 3.7 Plus lines, the Qwen-Plus-Character role-play models, and independent hosts are separate boundaries.", "strengths": [ "Alibaba documents the dated snapshot within a line of eight dated Qwen-Plus snapshots, separates it from the dynamically updated qwen-plus-latest alias, and states that the undated qwen-plus model is functionally equivalent to it", "1M context with switchable thinking and non-thinking modes, and function calling, structured output, and web search in the Chinese mainland deployment" ], "tradeoffs": [ "No downloadable weights exist under the Qwen-Plus name, so self-operation is unavailable", "Alibaba's capability tables disagree outside the Chinese mainland: the snapshot's list no supported features there, while the undated qwen-plus name that resolves to it lists structured output, prefix completion, and context caching", "Model Studio gives dated snapshots 30 days' retirement notice, against three months for mainline models" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 7, "overall": 4.38 }, "evidence": [ { "kind": "web", "label": "Official Model Studio qwen-plus page with the qwen-plus-2025-12-01 snapshot's capabilities, limits, and pricing", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen-plus", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio model lifecycle and updates page with the qwen-plus-2025-12-01 launch entry", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio pricing page listing the snapshot by deployment region", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio decommissioning policy: 30 days' notice for dated snapshots", "url": "https://www.alibabacloud.com/help/en/model-studio/model-depreciation", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Alibaba Cloud Model Studio Service Agreement", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Model Studio related agreements: governing terms index", "url": "https://help.aliyun.com/en/model-studio/related-agreements", "verified_at": "2026-09-28" } ], "metadata_verified_at": "2026-09-28", "verified_at": "2026-09-28" }, { "id": "model-alibaba-qwen-image-2-1", "source_id": "alibaba/qwen-image-2.1", "name": "Qwen-Image-2.1", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen-Image-2.1", "description": "Alibaba's Qwen-Image-2.1, a unified text-to-image generation and image editing model with a 7B-parameter visual generation component and native RGBA transparency, published as ungated open weights under the Qwen Research License, which permits non-commercial use only.", "model_type": "image_generation_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen-Image-2.1", "description": "Unified text-to-image generation and image editing model with native RGBA transparency and reference-image editing", "family": "qwen", "release_date": "2026-09-14", "last_updated": "2026-09-21", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 8192, "input": null, "output": 0 }, "reported_open_weights": true, "reported_license": "qwen-research", "links": [ { "label": "Blog", "url": "https://qwen.ai/blog?id=qwen-image-2.1", "type": "announcement" }, { "label": "GitHub", "url": "https://github.com/QwenLM/Qwen-Image-2.1", "type": "docs" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen-Image-2.1" } ] }, "licenses": [ "LicenseRef-Qwen-Research-License-Agreement" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The Qwen RESEARCH LICENSE AGREEMENT (release date September 20, 2026) is a single LICENSE file, byte-identical in the Hugging Face repository and the GitHub repository. It grants use, reproduction, distribution, derivative works, and modification for non-commercial purposes only, defining Non-Commercial as research or evaluation purposes only, and requires a separate commercial license requested from Alibaba for any commercial use. Redistribution requires a copy of the agreement, change notices, and a Notice file; any model created, trained, or improved using the Materials or their outputs must display 'Built with Qwen' or 'Improved using Qwen'; derivatives may not use Qwen as their primary name. The grant terminates if the licensee sues over IP in the Materials, and Alibaba may terminate on breach. The text contains no separate acceptable-use policy or field-of-use list beyond the non-commercial limit and export-control compliance, and the agreement is governed by Chinese law with exclusive jurisdiction in the Hangzhou courts.", "license_evidence": [ { "license_id": "LicenseRef-Qwen-Research-License-Agreement", "scope": "Qwen-Image-2.1 model weights, parameters, code, and documentation in the Qwen/Qwen-Image-2.1 Hugging Face repository and the QwenLM/Qwen-Image-2.1 repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen-Image-2.1 open-weight release announced September 20, 2026 and served from the ungated Qwen/Qwen-Image-2.1 repository. The hosted qwen-image-2.1-pro API model on Alibaba Cloud Model Studio, the Qwen-Image 2.0 and 3.0 API lines, earlier open Qwen-Image checkpoints, the Comfy-Org repackaged weights, FlagRelease chip-specific builds, and community quantizations are separate boundaries.", "strengths": [ "The weights ship ungated as Safetensors on Hugging Face and ModelScope with the licence file beside them and in the GitHub repository", "Day-0 documented paths through Diffusers, ComfyUI, vLLM-Omni, SGLang, and LightX2V, plus prebuilt FlagOS builds for eight non-NVIDIA chip platforms" ], "tradeoffs": [ "The licence permits only research or evaluation use, so any commercial deployment needs a separate license negotiated with Alibaba", "No first-party hosted endpoint serves this exact checkpoint, since the Model Studio API model is the separate qwen-image-2.1-pro, and no third-party host is documented" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.26 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/alibaba/qwen-image-2.1.toml", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen-Image-2.1 weights repository and model card", "url": "https://huggingface.co/Qwen/Qwen-Image-2.1", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Qwen Research License Agreement file in the weights repository", "url": "https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE", "verified_at": "2026-10-06" }, { "kind": "web", "label": "QwenLM release repository: September 20, 2026 release entry, ecosystem support, and licence", "url": "https://github.com/QwenLM/Qwen-Image-2.1", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Qwen announcement of September 20, 2026", "url": "https://qwen.ai/blog?id=qwen-image-2.1", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Alibaba Cloud Model Studio pricing page: the hosted image model is qwen-image-2.1-pro, a separate API model ID", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen2-5-72b-instruct", "source_id": null, "name": "Qwen2.5 72B Instruct", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct", "description": "Alibaba's 72.7-billion-parameter dense Qwen2.5 instruction model, published as downloadable weights under the custom Qwen License Agreement with a commercial-use authorization gate.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen2.5 72B Instruct", "description": "Instruction-tuned dense 72.7B-parameter Qwen2.5 language model", "family": "qwen", "release_date": "2024-09-19", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Qwen License Agreement", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct" } ] }, "licenses": [ "LicenseRef-Qwen-License-Agreement" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository LICENSE file is the Qwen License Agreement dated September 19, 2024, read in full. It grants a non-exclusive, worldwide, non-transferable, royalty-free limited license to use, reproduce, distribute, and modify the materials, requires recipients to receive the agreement and a Notice file with the Qwen attribution, and requires a request to Alibaba Cloud when a commercial product or service has more than 100 million monthly active users. Using the materials or their outputs to create, train, fine-tune, or improve a distributed AI model requires displaying 'Built with Qwen' or 'Improved using Qwen'. The license ends if the licensee sues Alibaba Cloud over infringement by the materials, requires the licensee to indemnify Alibaba Cloud, and is governed by Chinese law with Hangzhou courts, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Qwen-License-Agreement", "scope": "Qwen2.5-72B-Instruct model weights, parameters, code, and documentation, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/495f39366efef23836d0cfae4fbe635880d2be31/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen/Qwen2.5-72B-Instruct weight release of the September 2024 Qwen2.5 series at Hugging Face commit 495f39366efef23836d0cfae4fbe635880d2be31 (last modified 2025-01-12). The Base checkpoint, the other Qwen2.5 sizes, the Qwen2.5-Coder and Qwen2.5-VL lines, the Qwen3 successors, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 72B checkpoint carrying its governing license file beside the weights", "The release announcement states that the 72B size departs from Apache 2.0 and names the repository license file as the terms" ], "tradeoffs": [ "Commercial products above 100 million monthly active users need Alibaba Cloud's license, and training or improving a distributed model requires Qwen attribution", "No current exact-ID Model Studio serving page was found, and the license adds indemnity, a lawsuit-termination clause, and Chinese governing law" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.14 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen2.5 release announcement", "url": "https://qwenlm.github.io/blog/qwen2.5/", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen2-5-7b-instruct", "source_id": null, "name": "Qwen2.5 7B Instruct", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct", "description": "Alibaba's 7.61-billion-parameter dense Qwen2.5 instruction model, published as Apache-2.0 open weights.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen2.5 7B Instruct", "description": "Instruction-tuned dense 7.61B-parameter Qwen2.5 language model", "family": "qwen", "release_date": "2024-09-19", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, and the model-card metadata names apache-2.0 with a link to that file, matching the release announcement's statement that every size except 3B and 72B is Apache 2.0. It permits use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, and associated documentation in the Qwen/Qwen2.5-7B-Instruct repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/a09a35458c702b33eeacc393d103063234e8bc28/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen/Qwen2.5-7B-Instruct weight release of the September 2024 Qwen2.5 series at Hugging Face commit a09a35458c702b33eeacc393d103063234e8bc28 (last modified 2025-01-12). The Base checkpoint, the other Qwen2.5 sizes, the Qwen2.5-Coder and Qwen2.5-VL lines, the Qwen3 successors, and community quantizations remain separate boundaries.", "strengths": [ "Ungated dense checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "The release announcement states the Apache 2.0 terms for this size and points to the license files in each repository" ], "tradeoffs": [ "No current exact-ID Model Studio serving page was found, so served access must be arranged independently", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 8 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen2.5 release announcement", "url": "https://qwenlm.github.io/blog/qwen2.5/", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen2-5-coder-32b-instruct", "source_id": "alibaba/qwen2.5-coder-32b-instruct", "name": "Qwen2.5-Coder-32B-Instruct", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct", "description": "Alibaba's 32-billion-parameter Qwen2.5 coding release, published as Apache-2.0 open weights with Alibaba PAI deployment guidance.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen2.5-Coder-32B-Instruct", "description": "Open coding-focused Qwen model for code generation, repair, and repository reasoning", "family": "qwen", "release_date": "2024-11-12", "last_updated": "2024-11-12", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen2.5-Coder-32B-Instruct repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct/blob/381fc969f78efac66bc87ff7ddeadb7e73c218a7/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen2.5-Coder-32B-Instruct weight release of November 2024 at Hugging Face commit 381fc969f78efac66bc87ff7ddeadb7e73c218a7 (last modified 2025-01-12), with Alibaba PAI deployment guidance. The Base weights, the Qwen3 coder successors, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 32B coding checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "Alibaba PAI documentation uses this exact model for deployment, fine-tuning, and evaluation guidance" ], "tradeoffs": [ "No current exact-ID Model Studio serving page documents a managed endpoint for this line", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.8 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Alibaba PAI deployment guidance built on this exact model", "url": "https://www.alibabacloud.com/help/en/pai/use-cases/training-evaluation-compression-and-deployment-of-qwen2-5-coder-model", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen2.5-coder-32b-instruct.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen2-5-vl-72b-instruct", "source_id": "alibaba/qwen2-5-vl-72b-instruct", "name": "Qwen2.5-VL 72B Instruct", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct", "description": "Alibaba's 72-billion-parameter Qwen2.5 vision-language release, published as downloadable weights under the custom Qwen License Agreement with a commercial-use authorization gate.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen2.5-VL 72B Instruct", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "family": "qwen", "release_date": "2024-09", "last_updated": "2024-09", "knowledge_cutoff": "2024-04", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct" } ] }, "licenses": [ "LicenseRef-Qwen-License-Agreement" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository LICENSE file is the Qwen License Agreement dated September 19, 2024. It grants a non-exclusive, non-transferable, royalty-free license to use, reproduce, distribute, and modify the materials, but commercial products or services above 100 million monthly active users require Alibaba Cloud's express authorization, and training new models on the materials or their outputs requires a 'Built with Qwen' attribution, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Qwen-License-Agreement", "scope": "Qwen2.5-VL-72B-Instruct model weights, parameters, code, and documentation, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct/blob/89c86200743eec961a297729e7990e8f2ddbc4c5/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen2.5-VL-72B-Instruct weight release of January 2025 at Hugging Face commit 89c86200743eec961a297729e7990e8f2ddbc4c5 (last modified 2026-06-06), together with its exact-name commercial API listing. The smaller VL siblings, the Qwen3-VL successors, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 72B multimodal checkpoint carrying its governing license file beside the weights", "Alibaba's commercialization listing names qwen2.5-vl-72b-instruct as a managed-service model" ], "tradeoffs": [ "Commercial use above 100 million monthly active users needs Alibaba Cloud's express authorization, and training on the materials requires attribution", "No dedicated capability page documents limits for this line beyond the weight repository" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Alibaba Cloud commercialization listing naming qwen2.5-vl-72b-instruct", "url": "https://www.alibabacloud.com/en/notice/commercialization_launch_announcement_for_qwen25qwen25vl_opensource_free_trial_model_4d3", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen2-5-vl-72b-instruct.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-14b", "source_id": null, "name": "Qwen3 14B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-14B", "description": "Alibaba's 14.8-billion-parameter dense Qwen3 release that switches between thinking and non-thinking modes, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 14B", "description": "Dense 14.8B-parameter Qwen3 model with switchable thinking and non-thinking modes", "family": "qwen", "release_date": "2025-04-29", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-14B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, and the model-card metadata names apache-2.0 with a link to that file. It permits use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-14B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-14B/blob/40c069824f4251a91eefaf281ebe4c544efd3e18/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen/Qwen3-14B weight release of the April 2025 Qwen3 series, a hybrid model that switches between thinking and non-thinking modes, at Hugging Face commit 40c069824f4251a91eefaf281ebe4c544efd3e18 (last modified 2025-07-26), together with the hosted qwen3-14b API model. The Base checkpoint, the other Qwen3 sizes, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated dense checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio documentation names the exact qwen3-14b model ID as a served model" ], "tradeoffs": [ "Native context is 32K tokens and the 131K length needs a YaRN rope-scaling setting that the card says can affect shorter inputs", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-14B", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Model Studio qwen3-14b API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-14b", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen3 release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen3-235b-a22b-thinking-2507", "source_id": null, "name": "Qwen3 235B A22B Thinking 2507", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "description": "Alibaba's updated 235-billion-parameter Qwen3 mixture-of-experts model with 22B active parameters that supports only thinking mode, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 235B A22B Thinking 2507", "description": "Updated Qwen3 MoE model that supports only thinking mode, with 256K native context", "family": "qwen", "release_date": "2025-07-25", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2025 Alibaba Cloud, and the model-card metadata names apache-2.0 with a link to that file. It permits use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-235B-A22B-Thinking-2507 repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507/blob/6cbffae6d8e28b986a6b17bd36f42f9fa0f1f0a5/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen/Qwen3-235B-A22B-Thinking-2507 release, the July 2025 thinking-only update of Qwen3-235B-A22B, at Hugging Face commit 6cbffae6d8e28b986a6b17bd36f42f9fa0f1f0a5 (last modified 2025-08-17), together with the hosted qwen3-235b-a22b-thinking-2507 API model. The original hybrid Qwen3-235B-A22B, the Instruct-2507 variant, FP8 and other quantized repackagings, and later Qwen generations remain separate boundaries.", "strengths": [ "Ungated Safetensors shards with a standard Apache-2.0 license file are published directly by the developer", "First-party Model Studio serving under the exact qwen3-235b-a22b-thinking-2507 model ID" ], "tradeoffs": [ "The 235B full-precision artifact is very large, so self-hosting needs substantial multi-GPU capacity", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "overall": 7.88, "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9 }, "evidence": [ { "kind": "web", "label": "Official model card, weight repository, and deployment guidance", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Model Studio qwen3-235b-a22b-thinking-2507 API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-235b-a22b-thinking-2507", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen3 repository with dated release notes", "url": "https://github.com/QwenLM/Qwen3", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen3-235b-a22b", "source_id": "alibaba/qwen3-235b-a22b", "name": "Qwen3 235B-A22B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B", "description": "Alibaba's 235-billion-parameter Mixture-of-Experts Qwen3 flagship with 22B active parameters, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 235B-A22B", "description": "Large open Qwen MoE for multilingual reasoning, coding, and tool use", "family": "qwen", "release_date": "2025-04", "last_updated": "2025-04", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-235B-A22B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/8efa61729e24bd65b1d152b5ab5409052aa80e65/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3-235B-A22B weight release of the April 2025 Qwen3 series at Hugging Face commit 8efa61729e24bd65b1d152b5ab5409052aa80e65 (last modified 2025-07-26), together with the hosted qwen3-235b-a22b API model. The Base weights, 2507 Instruct and Thinking releases, NVFP4 and FP8 variants, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated flagship MoE checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio serving under the exact qwen3-235b-a22b model ID" ], "tradeoffs": [ "235B total parameters need multi-GPU serving, putting self-operation out of reach for most operators", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 9, "overall": 7.76 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3-235b-a22b API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-235b-a22b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-235b-a22b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-30b-a3b", "source_id": "alibaba/qwen3-30b-a3b", "name": "Qwen3 30B A3B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B", "description": "Alibaba's 30-billion-parameter Mixture-of-Experts Qwen3 release with 3B active parameters, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 30B A3B", "description": "Sparse MoE Qwen model with 3B active parameters for efficient chat and reasoning", "family": "qwen", "release_date": "2025-04-28", "last_updated": "2025-04-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-30B-A3B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B/blob/ad44e777bcd18fa416d9da3bd8f70d33ebb85d39/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3-30B-A3B weight release of the April 2025 Qwen3 series at Hugging Face commit ad44e777bcd18fa416d9da3bd8f70d33ebb85d39 (last modified 2025-07-26), together with the hosted qwen3-30b-a3b API model. The Base and 2507 Instruct and Thinking successors, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated MoE checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio serving under the exact qwen3-30b-a3b model ID" ], "tradeoffs": [ "The 2507 Instruct and Thinking successors blur the line's forward identity", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 9, "overall": 8.16 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3-30b-a3b API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-30b-a3b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-30b-a3b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-30b-a3b-instruct-2507", "source_id": null, "name": "Qwen3 30B A3B Instruct 2507", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507", "description": "Alibaba's updated 30.5-billion-parameter Qwen3 mixture-of-experts model with 3.3B active parameters that supports only non-thinking mode, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 30B A3B Instruct 2507", "description": "Updated Qwen3 MoE model that supports only non-thinking mode, with 256K native context", "family": "qwen", "release_date": "2025-07-30", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2025 Alibaba Cloud, and the model-card metadata names apache-2.0 with a link to that file. It permits use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-30B-A3B-Instruct-2507 repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/0d7cf23991f47feeb3a57ecb4c9cee8ea4a17bfe/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen/Qwen3-30B-A3B-Instruct-2507 release, the July 2025 non-thinking update of Qwen3-30B-A3B, at Hugging Face commit 0d7cf23991f47feeb3a57ecb4c9cee8ea4a17bfe (last modified 2025-09-17), together with the hosted qwen3-30b-a3b-instruct-2507 API model. The original hybrid Qwen3-30B-A3B, the Thinking-2507 variant, FP8 and other quantized repackagings, and later Qwen generations remain separate boundaries.", "strengths": [ "Ungated Safetensors shards with a standard Apache-2.0 license file are published directly by the developer", "First-party Model Studio serving under the exact qwen3-30b-a3b-instruct-2507 model ID" ], "tradeoffs": [ "The model card replaces the thinking switch of the original release, so settings written for qwen3-30b-a3b do not carry over", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "overall": 8.28, "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9 }, "evidence": [ { "kind": "web", "label": "Official model card, weight repository, and deployment guidance", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Model Studio qwen3-30b-a3b-instruct-2507 API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-30b-a3b-instruct-2507", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen3 repository with dated release notes", "url": "https://github.com/QwenLM/Qwen3", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen3-30b-a3b-thinking-2507", "source_id": null, "name": "Qwen3 30B A3B Thinking 2507", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507", "description": "Alibaba's updated 30.5-billion-parameter Qwen3 mixture-of-experts model with 3.3B active parameters that supports only thinking mode, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 30B A3B Thinking 2507", "description": "Updated Qwen3 MoE model that supports only thinking mode, with 256K native context", "family": "qwen", "release_date": "2025-07-31", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2025 Alibaba Cloud, and the model-card metadata names apache-2.0 with a link to that file. It permits use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-30B-A3B-Thinking-2507 repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/blob/144afc2f379b542fdd4e85a1fcd5e1f79112d95d/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen/Qwen3-30B-A3B-Thinking-2507 release, the July 2025 thinking-only update of Qwen3-30B-A3B, at Hugging Face commit 144afc2f379b542fdd4e85a1fcd5e1f79112d95d (last modified 2025-08-17), together with the hosted qwen3-30b-a3b-thinking-2507 API model. The original hybrid Qwen3-30B-A3B, the Instruct-2507 variant, FP8 and other quantized repackagings, and later Qwen generations remain separate boundaries.", "strengths": [ "Ungated Safetensors shards with a standard Apache-2.0 license file are published directly by the developer", "First-party Model Studio serving under the exact qwen3-30b-a3b-thinking-2507 model ID" ], "tradeoffs": [ "Thinking is always on and the chat template inserts the opening think tag, so output parsing differs from the hybrid release", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "overall": 8.28, "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9 }, "evidence": [ { "kind": "web", "label": "Official model card, weight repository, and deployment guidance", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Model Studio qwen3-30b-a3b-thinking-2507 API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-30b-a3b-thinking-2507", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen3 repository with dated release notes", "url": "https://github.com/QwenLM/Qwen3", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen3-32b", "source_id": "alibaba/qwen3-32b", "name": "Qwen3 32B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-32B", "description": "Alibaba's 32-billion-parameter dense Qwen3 release for agentic and general work, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 32B", "description": "Dense open Qwen model for self-hosted chat, reasoning, and coding", "family": "qwen", "release_date": "2025-04", "last_updated": "2025-04", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-32B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-32B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-32B/blob/9216db5781bf21249d130ec9da846c4624c16137/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3-32B weight release of the April 2025 Qwen3 series at Hugging Face commit 9216db5781bf21249d130ec9da846c4624c16137 (last modified 2025-07-26), together with the hosted qwen3-32b API model. The Qwen3-32B-Base weights, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated weight repository carrying the verbatim Apache-2.0 license file beside the checkpoint", "First-party Model Studio serving under the exact qwen3-32b model ID" ], "tradeoffs": [ "Dense 32B still needs a high-memory single GPU or quantization for smaller self-hosted setups", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-32B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3-32b API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-32b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-32b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-8b", "source_id": null, "name": "Qwen3 8B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-8B", "description": "Alibaba's 8.2-billion-parameter dense Qwen3 release that switches between thinking and non-thinking modes, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3 8B", "description": "Dense 8.2B-parameter Qwen3 model with switchable thinking and non-thinking modes", "family": "qwen", "release_date": "2025-04-29", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-8B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, and the model-card metadata names apache-2.0 with a link to that file. It permits use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-8B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-8B/blob/b968826d9c46dd6066d109eabc6255188de91218/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Qwen/Qwen3-8B weight release of the April 2025 Qwen3 series, a hybrid model that switches between thinking and non-thinking modes, at Hugging Face commit b968826d9c46dd6066d109eabc6255188de91218 (last modified 2025-07-26), together with the hosted qwen3-8b API model. The Base checkpoint, the other Qwen3 sizes, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated dense checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio documentation names the exact qwen3-8b model ID as a served model" ], "tradeoffs": [ "Native context is 32K tokens and the 131K length needs a YaRN rope-scaling setting that the card says can affect shorter inputs", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-8B", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Model Studio qwen3-8b API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-8b", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Qwen3 release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen3-embedding-8b", "source_id": "alibaba/qwen3-embedding-8b", "name": "Qwen3 Embedding 8B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-Embedding-8B", "description": "Alibaba's June 2025 Qwen3 Embedding release at 8B parameters with 4,096 dimensions, multilingual retrieval across more than one hundred languages, and Matryoshka-style flexible vectors, published under Apache 2.0 with weights on the Qwen Hugging Face organization. No developer API serves this checkpoint. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "embedding_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen3 Embedding 8B", "description": "Multilingual text embedding model for retrieval, code search, classification, and clustering", "family": "qwen", "release_date": "2025-06-05", "last_updated": "2025-06-05", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text", "embedding" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 32768, "input": null, "output": 0 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Embedding-8B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The Qwen team's announcement states the Qwen3 Embedding series is open-sourced under Apache 2.0, and the Qwen/Qwen3-Embedding-8B repository carries an Apache License, Version 2.0 LICENSE file and declares license apache-2.0. No separate model-use terms were found on the pages read.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Qwen3-Embedding-8B weights published in Alibaba's Qwen Hugging Face repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-Embedding-8B/blob/1d8ad4ca9b3dd8059ad90a75d4983776a23d44af/LICENSE", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the Qwen3-Embedding-8B checkpoint. The 0.6B and 4B siblings, the Qwen3-Reranker line, the Qwen3 language models the series is built on, and ModelScope mirror pages are separate boundaries.", "strengths": [ "Apache 2.0 grant stated in the release announcement and carried in a LICENSE file inside the weight repository", "Top-ranked multilingual retrieval at release with flexible vector dimensions and user-defined instructions in an 8B checkpoint" ], "tradeoffs": [ "No developer API or managed host serves this checkpoint, so running it is entirely self-operated", "The 8B checkpoint needs accelerator-class hardware where the 0.6B sibling runs leaner, so sizing is workload-dependent" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.14 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/59aa52dbc09727fa6ead433b0285970ebc05ba82/models/alibaba/qwen3-embedding-8b.toml", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Qwen3 Embedding announcement: June 5, 2025 series release under Apache 2.0", "url": "https://qwenlm.github.io/blog/qwen3-embedding/", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Qwen Qwen3-Embedding-8B weight repository: apache-2.0 licence", "url": "https://huggingface.co/Qwen/Qwen3-Embedding-8B", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/Qwen/Qwen3-Embedding-8B", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-alibaba-qwen3-max-2026-01-23", "source_id": null, "name": "Qwen3 Max 2026-01-23", "developer": "Alibaba", "url": "https://www.alibabacloud.com/help/en/model-studio/model-qwen3-max", "description": "Alibaba's January 23, 2026 snapshot of Qwen3-Max, a text model that integrates thinking and non-thinking modes and adds built-in web search, web extraction, and code interpreter tools in thinking mode, served through Alibaba Cloud Model Studio with 262K context and no downloadable weights.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Qwen3 Max 2026-01-23", "description": "Qwen3-Max snapshot integrating thinking and non-thinking modes, with web search, web extraction, and code interpreter tools in thinking mode", "family": "qwen", "release_date": "2026-01-23", "last_updated": "2026-01-23", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 262144, "input": 258048, "output": 65536 }, "reported_open_weights": null, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Alibaba-Model-Studio-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The qwen3-max-2026-01-23 snapshot is served as a Model Studio API model under the Alibaba Cloud Model Studio Service Agreement, effective September 2, 2026, which Model Studio's related-agreements page lists as governing the service. It is a paid-service contract that bars reselling the service or its models and using them or their outputs to train or develop competing products without written permission, and prohibits reverse engineering or trying to discover the service's source code. It grants no licence to model weights for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Alibaba-Model-Studio-Agreement", "scope": "Business and developer access to the qwen3-max-2026-01-23 snapshot through Alibaba Cloud Model Studio API services", "kind": "web_terms", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" } ], "access_boundary": "This record represents the Qwen3-Max line at its current pinned snapshot, qwen3-max-2026-01-23, which Alibaba Cloud Model Studio serves with 262,144-token context and to which the undated qwen3-max model currently resolves. The earlier qwen3-max-2025-09-23 and qwen3-max-preview versions, the Qwen3.6, 3.7, and 3.8 Max lines, open-weight Qwen3 releases, and independent hosts are separate boundaries.", "strengths": [ "Alibaba documents the dated snapshot with its own capability, context-limit, rate-limit, and pricing tables, and states that the undated qwen3-max model is functionally equivalent to it", "Function calling and structured output alongside switchable thinking, with web search, web extraction, and a code interpreter available in thinking mode" ], "tradeoffs": [ "No downloadable weights exist under the Qwen3-Max name, so self-operation is unavailable", "Model Studio gives dated snapshots 30 days' retirement notice, against three months for mainline models" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Model Studio qwen3-max page with the qwen3-max-2026-01-23 snapshot's capabilities, limits, and pricing", "url": "https://www.alibabacloud.com/help/en/model-studio/model-qwen3-max", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio model lifecycle and updates page with the qwen3-max-2026-01-23 launch entry", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio pricing page listing the snapshot by deployment region", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio decommissioning policy: 30 days' notice for dated snapshots", "url": "https://www.alibabacloud.com/help/en/model-studio/model-depreciation", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Alibaba Cloud Model Studio Service Agreement", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Model Studio related agreements: governing terms index", "url": "https://help.aliyun.com/en/model-studio/related-agreements", "verified_at": "2026-09-28" } ], "metadata_verified_at": "2026-09-28", "verified_at": "2026-09-28" }, { "id": "model-alibaba-qwen3-coder-30b-a3b-instruct", "source_id": "alibaba/qwen3-coder-30b-a3b-instruct", "name": "Qwen3-Coder 30B-A3B Instruct", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct", "description": "Alibaba's 30-billion-parameter Mixture-of-Experts coding release with 3B active parameters for agentic coding workflows, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3-Coder 30B-A3B Instruct", "description": "Smaller Qwen coder for efficient local agents and repo-level fixes", "family": "qwen", "release_date": "2025-04", "last_updated": "2025-04", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-Coder-30B-A3B-Instruct repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/b2cff646eb4bb1d68355c01b18ae02e7cf42d120/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3-Coder-30B-A3B-Instruct weight release of July 2025 at Hugging Face commit b2cff646eb4bb1d68355c01b18ae02e7cf42d120 (last modified 2025-12-03), together with the hosted qwen3-coder-30b-a3b-instruct API model. The Base weights, the 480B flagship sibling, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated MoE coding checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio serving under the exact qwen3-coder-30b-a3b-instruct model ID" ], "tradeoffs": [ "Version identity rests on repository commits with no stated retirement posture", "Access, quotas, and lifecycle remain controlled by the managed API rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3-coder-30b-a3b-instruct API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-coder-30b-a3b-instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-coder-30b-a3b-instruct.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-coder-480b-a35b-instruct", "source_id": "alibaba/qwen3-coder-480b-a35b-instruct", "name": "Qwen3-Coder 480B-A35B Instruct", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct", "description": "Alibaba's 480-billion-parameter Mixture-of-Experts flagship coding release with 35B active parameters, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3-Coder 480B-A35B Instruct", "description": "Open Qwen coding heavyweight for repository reasoning and agentic engineering", "family": "qwen", "release_date": "2025-04", "last_updated": "2025-04", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-Coder-480B-A35B-Instruct repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3-Coder-480B-A35B-Instruct weight release of July 2025 at Hugging Face commit 9d90cf8fca1bf7b7acca42d3fc9ae694a2194069 (last modified 2025-08-21), together with the hosted qwen3-coder-480b-a35b-instruct API model. The Base weights, the 30B sibling, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 480B flagship checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio serving under the exact qwen3-coder-480b-a35b-instruct model ID" ], "tradeoffs": [ "480B total parameters need multi-GPU-cluster serving, putting self-operation out of reach for most operators", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 4, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.48 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3-coder-480b-a35b-instruct API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-coder-480b-a35b-instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-coder-480b-a35b-instruct.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-coder-next", "source_id": "alibaba/qwen3-coder-next", "name": "Qwen3-Coder-Next", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next", "description": "Alibaba's Qwen3-Coder-Next, an 80-billion-parameter coding and agentic tool-use MoE release with 3B active parameters and 256K context, published as open weights under Apache-2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Qwen3 Coder Next", "description": "Open-weight Qwen coding model for agents, repository edits, and multi-turn tool use", "family": "qwen", "release_date": "2026-02-03", "last_updated": "2026-02-03", "knowledge_cutoff": "2025-09", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no separate LICENSE file; Apache-2.0 rests on the pinned model-card metadata and the Qwen3 family's Apache-2.0 distribution practice. The license grant is treated as the same Apache-2.0 surface used by the sibling Qwen3-Coder and Qwen3-Next releases.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Qwen3-Coder-Next repository weights, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Qwen3-Coder-Next weight release, a coding and agentic-tool-use model with 80B total and 3B active parameters published in the Qwen organization. The Qwen3-Coder 30B and 480B checkpoints, the general Qwen3-Next 80B releases, later Qwen generations, hosted Model Studio lines, and community quantizations are separate boundaries.", "strengths": [ "Ungated 80B/3B-active MoE checkpoints with a dedicated coding-agent and tool-use focus", "256K context, a published technical report, and documented vLLM and SGLang serving paths" ], "tradeoffs": [ "The repository ships no LICENSE file, so Apache-2.0 rests on pinned model-card metadata", "The roughly 159 GB checkpoint set is heavy for local deployment despite the small active parameter set" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.22 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-coder-next.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-alibaba-qwen3-next-80b-a3b-thinking", "source_id": "alibaba/qwen3-next-80b-a3b-thinking", "name": "Qwen3-Next 80B-A3B (Thinking)", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking", "description": "Alibaba's 80-billion-parameter Qwen3-Next thinking release with enforced reasoning, published as a distinct Apache-2.0 open-weights checkpoint with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3-Next 80B-A3B (Thinking)", "description": "Efficient Qwen thinking model for local reasoning, math, and coding agents", "family": "qwen", "release_date": "2025-09", "last_updated": "2025-09", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-Next-80B-A3B-Thinking repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking/blob/e502dd4100cc68c0de57643fd4317ec93a128670/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3-Next-80B-A3B-Thinking weight release of September 2025 at Hugging Face commit e502dd4100cc68c0de57643fd4317ec93a128670 (last modified 2025-09-15), a distinct post-trained checkpoint from the Instruct sibling, together with the hosted qwen3-next-80b-a3b-thinking API model. The Base weights, the Instruct sibling, the FP8 variant, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated thinking checkpoint carrying the verbatim Apache-2.0 license file beside the weights, in a distinct repository from the Instruct sibling", "First-party Model Studio serving under the exact qwen3-next-80b-a3b-thinking model ID" ], "tradeoffs": [ "Thinking workloads carry heavier context and output budgets than the Instruct sibling", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3-next-80b-a3b-thinking API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-next-80b-a3b-thinking", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-next-80b-a3b-thinking.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-next-80b-a3b-instruct", "source_id": "alibaba/qwen3-next-80b-a3b-instruct", "name": "Qwen3-Next 80B-A3B Instruct", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "description": "Alibaba's 80-billion-parameter Qwen3-Next instruct release for non-thinking agentic work, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3-Next 80B-A3B Instruct", "description": "Qwen instruction model for multilingual chat, reasoning, and tool use", "family": "qwen", "release_date": "2025-09", "last_updated": "2025-09", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3-Next-80B-A3B-Instruct repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct/blob/9c7f2fbe84465e40164a94cc16cd30b6999b0cc7/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3-Next-80B-A3B-Instruct weight release of September 2025 at Hugging Face commit 9c7f2fbe84465e40164a94cc16cd30b6999b0cc7 (last modified 2025-09-17), together with the hosted qwen3-next-80b-a3b-instruct API model. The Base weights, the Thinking sibling release, the FP8 variant, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 80B checkpoint carrying the verbatim Apache-2.0 license file beside the weights, plus an official FP8 sibling", "First-party Model Studio serving under the exact qwen3-next-80b-a3b-instruct model ID" ], "tradeoffs": [ "163 GB of weights need multi-GPU serving for full-precision self-operation", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3-next-80b-a3b-instruct API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-next-80b-a3b-instruct", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3-next-80b-a3b-instruct.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-5-122b-a10b", "source_id": "alibaba/qwen3.5-122b-a10b", "name": "Qwen3.5 122B-A10B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "description": "Alibaba's 122-billion-parameter Mixture-of-Experts Qwen3.5 release with 10B active parameters, published as Apache-2.0 open weights with Model Studio serving.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3.5 122B-A10B", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "family": "qwen", "release_date": "2026-02-23", "last_updated": "2026-02-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.5-122B-A10B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B/blob/dc4d348443bc740c68e2d77492492c11606384d5/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3.5-122B-A10B weight release of February 2026 at Hugging Face commit dc4d348443bc740c68e2d77492492c11606384d5 (last modified 2026-04-24), together with its Model Studio serving row. Smaller and larger Qwen3.5 siblings, the hosted Flash counterpart line, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 122B MoE checkpoint carrying the verbatim Apache-2.0 license file from day one", "Model Studio lists qwen3.5-122b-a10b as a served model" ], "tradeoffs": [ "122B total parameters need tensor-parallel serving across multiple GPUs", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.22 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio model listing naming qwen3.5-122b-a10b", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.5-122b-a10b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-5-27b", "source_id": "alibaba/qwen3.5-27b", "name": "Qwen3.5 27B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "description": "Alibaba's 27-billion-parameter dense Qwen3.5 release with a vision encoder, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3.5 27B", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "family": "qwen", "release_date": "2026-02-23", "last_updated": "2026-02-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-27B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.5-27B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.5-27B/blob/fc05daec18b0a78c049392ed2e771dde82bdf654/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3.5-27B weight release of February 2026 at Hugging Face commit fc05daec18b0a78c049392ed2e771dde82bdf654 (last modified 2026-04-24), together with the hosted qwen3.5-27b API model. Smaller and larger Qwen3.5 siblings, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 27B dense checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio serving under the exact qwen3.5-27b model ID" ], "tradeoffs": [ "The license file arrived about two months after the weight release", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 6, "documentation_provenance": 9, "overall": 8.3 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3.5-27b API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-27b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.5-27b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-5-35b-a3b", "source_id": "alibaba/qwen3.5-35b-a3b", "name": "Qwen3.5 35B-A3B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "description": "Alibaba's 35-billion-parameter Mixture-of-Experts Qwen3.5 release with 3B active parameters, published as Apache-2.0 open weights with Model Studio serving.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3.5 35B-A3B", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "family": "qwen", "release_date": "2026-02-23", "last_updated": "2026-02-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.5-35B-A3B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/59d61f3ce65a6d9863b86d2e96597125219dc754/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3.5-35B-A3B weight release of February 2026 at Hugging Face commit 59d61f3ce65a6d9863b86d2e96597125219dc754 (last modified 2026-04-24), together with its Model Studio serving row. Smaller and larger Qwen3.5 siblings, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 35B MoE checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "Model Studio lists qwen3.5-35b-a3b as a served model" ], "tradeoffs": [ "Version identity rests on repository commits with no stated retirement posture", "Access, quotas, and lifecycle remain controlled by the managed API rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.42 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio model listing naming qwen3.5-35b-a3b", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.5-35b-a3b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-5-397b-a17b", "source_id": "alibaba/qwen3.5-397b-a17b", "name": "Qwen3.5 397B-A17B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "description": "Alibaba's 397-billion-parameter Mixture-of-Experts Qwen3.5 flagship with 17B active parameters, published as Apache-2.0 open weights with Model Studio serving.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3.5 397B-A17B", "description": "Large open Qwen multimodal MoE for visual agents and long technical tasks", "family": "qwen", "release_date": "2026-02-15", "last_updated": "2026-02-15", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.5-397B-A17B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/blob/8472618112abcbd45acbcdc58436aff4233c23f7/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3.5-397B-A17B weight release of February 2026 at Hugging Face commit 8472618112abcbd45acbcdc58436aff4233c23f7 (last modified 2026-04-24), together with its Model Studio serving row. Smaller Qwen3.5 siblings, the hosted Plus counterpart line, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 397B flagship checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "Model Studio lists qwen3.5-397b-a17b as a served model" ], "tradeoffs": [ "397B total parameters need cluster-class serving, putting self-operation out of reach for most operators", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.02 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio model listing naming qwen3.5-397b-a17b", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.5-397b-a17b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-5-9b", "source_id": "alibaba/qwen3.5-9b", "name": "Qwen3.5 9B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.5-9B", "description": "Alibaba's 9-billion-parameter dense Qwen3.5 release with a vision encoder, published as Apache-2.0 open weights for single-GPU self-operation.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen3.5 9B", "description": "Qwen instruction model for multilingual chat, reasoning, and tool use", "family": "qwen", "release_date": "2026-02-23", "last_updated": "2026-02-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-9B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.5-9B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.5-9B/blob/c202236235762e1c871ad0ccb60c8ee5ba337b9a/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3.5-9B weight release of February 2026 at Hugging Face commit c202236235762e1c871ad0ccb60c8ee5ba337b9a (last modified 2026-03-02), with in-repository SGLang, vLLM, and Transformers serving recipes. Smaller Qwen3.5 dense siblings, later Qwen generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 9B dense checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "Single-GPU self-operation with documented SGLang and vLLM serving recipes" ], "tradeoffs": [ "No Model Studio model page documents a managed endpoint for this exact ID", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.34 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and serving recipes", "url": "https://huggingface.co/Qwen/Qwen3.5-9B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.5-9b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-5-flash-2026-02-23", "source_id": null, "name": "Qwen3.5 Flash 2026-02-23", "developer": "Alibaba", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-flash", "description": "Alibaba's February 23, 2026 snapshot of Qwen3.5-Flash, a native vision-language model that takes text, image, and video input with 1M context, served through Alibaba Cloud Model Studio as the hosted counterpart of the open Qwen3.5-35B-A3B weights, with no downloadable weights under its own name.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Qwen3.5 Flash 2026-02-23", "description": "Qwen3.5 native vision-language Flash snapshot built on a hybrid linear-attention and sparse mixture-of-experts architecture", "family": "qwen", "release_date": "2026-02-23", "last_updated": "2026-02-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 1000000, "input": 991808, "output": 65536 }, "reported_open_weights": null, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Alibaba-Model-Studio-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The qwen3.5-flash-2026-02-23 snapshot is served as a Model Studio API model under the Alibaba Cloud Model Studio Service Agreement, effective September 2, 2026, which Model Studio's related-agreements page lists as governing the service. It is a paid-service contract that bars reselling the service or its models and using them or their outputs to train or develop competing products without written permission, and prohibits reverse engineering or trying to discover the service's source code. It grants no licence to model weights for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Alibaba-Model-Studio-Agreement", "scope": "Business and developer access to the qwen3.5-flash-2026-02-23 snapshot through Alibaba Cloud Model Studio API services", "kind": "web_terms", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" } ], "access_boundary": "This record represents the Qwen3.5-Flash line at its only dated snapshot, qwen3.5-flash-2026-02-23, which Alibaba Cloud Model Studio serves with text, image, and video input and 1M context and to which the undated qwen3.5-flash model resolves. The Qwen3.5-35B-A3B open weights that Alibaba calls Qwen3.5-Flash's open counterpart, the Qwen3.5-Plus line, later Qwen generations, and independent hosts are separate boundaries.", "strengths": [ "Alibaba documents the dated snapshot with its own capability, context-limit, and pricing tables, states that the undated qwen3.5-flash model is functionally equivalent to it, and names the open weights it corresponds to", "Text, image, and video input with 1M context, function calling, structured output, and web search, with Model Studio fine-tuning documented for the Chinese mainland deployment" ], "tradeoffs": [ "No downloadable weights exist under the Qwen3.5-Flash name; the corresponding Qwen3.5-35B-A3B open weights are a separate release that Alibaba says lacks hosted production features such as the 1M default context and built-in tools", "Model Studio gives dated snapshots 30 days' retirement notice, against three months for mainline models" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Model Studio qwen3.5-flash page with the qwen3.5-flash-2026-02-23 snapshot's capabilities, limits, and pricing", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-flash", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Qwen3.5-35B-A3B model card naming Qwen3.5-Flash as its hosted counterpart", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio model lifecycle and updates page with the qwen3.5-flash-2026-02-23 launch entry", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio pricing page listing the snapshot by deployment region", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio decommissioning policy: 30 days' notice for dated snapshots", "url": "https://www.alibabacloud.com/help/en/model-studio/model-depreciation", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Alibaba Cloud Model Studio Service Agreement", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Model Studio related agreements: governing terms index", "url": "https://help.aliyun.com/en/model-studio/related-agreements", "verified_at": "2026-09-28" } ], "metadata_verified_at": "2026-09-28", "verified_at": "2026-09-28" }, { "id": "model-alibaba-qwen3-5-plus-2026-04-20", "source_id": null, "name": "Qwen3.5 Plus 2026-04-20", "developer": "Alibaba", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-plus", "description": "Alibaba's April 20, 2026 snapshot of Qwen3.5-Plus, a native vision-language model that takes text, image, and video input with 1M context, improved over the February snapshot for agentic coding and served through Alibaba Cloud Model Studio with no downloadable weights under its own name.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Qwen3.5 Plus 2026-04-20", "description": "Qwen3.5 native vision-language Plus snapshot with improved agentic coding and inference speed over the February 15, 2026 snapshot", "family": "qwen", "release_date": "2026-04-23", "last_updated": "2026-04-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 1000000, "input": 991808, "output": 65536 }, "reported_open_weights": null, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Alibaba-Model-Studio-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The qwen3.5-plus-2026-04-20 snapshot is served as a Model Studio API model under the Alibaba Cloud Model Studio Service Agreement, effective September 2, 2026, which Model Studio's related-agreements page lists as governing the service. It is a paid-service contract that bars reselling the service or its models and using them or their outputs to train or develop competing products without written permission, and prohibits reverse engineering or trying to discover the service's source code. It grants no licence to model weights for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Alibaba-Model-Studio-Agreement", "scope": "Business and developer access to the qwen3.5-plus-2026-04-20 snapshot through Alibaba Cloud Model Studio API services", "kind": "web_terms", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" } ], "access_boundary": "This record represents the Qwen3.5-Plus line at its newest dated snapshot, qwen3.5-plus-2026-04-20, which Alibaba Cloud Model Studio serves in its Singapore and Beijing regions with text, image, and video input and 1M context. The undated qwen3.5-plus model still resolves to the February 15, 2026 snapshot, which this record does not cover. That snapshot, the Qwen3.5-397B-A17B open weights that Alibaba calls Qwen3.5-Plus's open counterpart, the Qwen3.5-Flash line, later Qwen generations, and independent hosts are separate boundaries.", "strengths": [ "Alibaba documents the dated snapshot with its own capability, context-limit, and pricing tables and names its agentic-coding and speed gains over the February 15 snapshot", "Text, image, and video input with 1M context, function calling, structured output, web search, and context caching" ], "tradeoffs": [ "No downloadable weights exist under the Qwen3.5-Plus name; the corresponding Qwen3.5-397B-A17B open weights are a separate release that Alibaba says lacks hosted production features such as the 1M default context and built-in tools", "Served only in the Singapore and Beijing regions while the undated qwen3.5-plus name still resolves to the February snapshot, and dated snapshots get 30 days' retirement notice" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Model Studio qwen3.5-plus page with the qwen3.5-plus-2026-04-20 snapshot's capabilities, limits, and pricing", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-plus", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Qwen3.5-397B-A17B model card naming Qwen3.5-Plus as its hosted counterpart", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio model lifecycle and updates page with the qwen3.5-plus-2026-04-20 launch entry", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio pricing page listing the snapshot by deployment region", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Model Studio decommissioning policy: 30 days' notice for dated snapshots", "url": "https://www.alibabacloud.com/help/en/model-studio/model-depreciation", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Alibaba Cloud Model Studio Service Agreement", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Model Studio related agreements: governing terms index", "url": "https://help.aliyun.com/en/model-studio/related-agreements", "verified_at": "2026-09-28" } ], "metadata_verified_at": "2026-09-28", "verified_at": "2026-09-28" }, { "id": "model-alibaba-qwen3-6-27b", "source_id": "alibaba/qwen3.6-27b", "name": "Qwen3.6 27B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "description": "Alibaba's 27-billion-parameter dense Qwen3.6 release with native multimodal input, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3.6 27B", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "family": "qwen", "release_date": "2026-04-22", "last_updated": "2026-04-22", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.6-27B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.6-27B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.6-27B/blob/6a9e13bd6fc8f0983b9b99948120bc37f49c13e9/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3.6-27B weight release of April 2026 at Hugging Face commit 6a9e13bd6fc8f0983b9b99948120bc37f49c13e9 (last modified 2026-04-24), together with the hosted qwen3.6-27b API model. The 35B MoE sibling, earlier Qwen generations, later Qwen lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 27B dense checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio serving under the exact qwen3.6-27b model ID with image, text, and video input" ], "tradeoffs": [ "Version identity rests on repository commits with no stated retirement posture", "Access, quotas, and lifecycle remain controlled by the managed API rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3.6-27b API documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-6-27b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.6-27b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-6-35b-a3b", "source_id": "alibaba/qwen3.6-35b-a3b", "name": "Qwen3.6 35B-A3B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "description": "Alibaba's 35-billion-parameter Mixture-of-Experts Qwen3.6 release with 3B active parameters, published as Apache-2.0 open weights with a hosted Model Studio API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Qwen3.6 35B-A3B", "description": "Open multimodal Qwen MoE for local agents that need vision, audio, and code", "family": "qwen", "release_date": "2026-04-17", "last_updated": "2026-04-17", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.6-35B-A3B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B/blob/995ad96eacd98c81ed38be0c5b274b04031597b0/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/Qwen3.6-35B-A3B weight release of April 2026 at Hugging Face commit 995ad96eacd98c81ed38be0c5b274b04031597b0 (last modified 2026-04-24), together with the hosted qwen3.6-35b-a3b API model. The 27B dense sibling, earlier Qwen generations, later Qwen lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 35B MoE checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "First-party Model Studio serving under the exact qwen3.6-35b-a3b model ID" ], "tradeoffs": [ "Version identity rests on repository commits with no stated retirement posture", "Access, quotas, and lifecycle remain controlled by the managed API rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.42 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Model Studio qwen3.6-35b-a3b API documentation", "url": "https://docs.modelstudio.console.alibabacloud.com/en/model-studio/qwen3-6-35b-a3b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.6-35b-a3b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-8-27b", "source_id": "alibaba/qwen3.8-27b", "name": "Qwen3.8 27B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "description": "Alibaba's 27-billion-parameter dense vision-language model for coding, office productivity, and long-horizon agentic work, published as Apache-2.0 open weights and served as the hosted qwen3.8-27b API model.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Qwen3.8 27B", "description": "Dense 27B vision-language model for coding, agent tasks, and image and video understanding", "family": "qwen", "release_date": "2026-08-14", "last_updated": "2026-08-14", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.8-27B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The downloadable weights carry the verbatim Apache License 2.0, appendix completed as Copyright 2026 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license — unlike the custom license on the 2.4T Max sibling. Hosted qwen3.8-27b API use is governed separately by QwenCloud and Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.8-27B repository; does not cover the hosted qwen3.8-27b API service", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.8-27B/blob/main/LICENSE", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the Qwen/Qwen3.8-27B open-weights checkpoint plus official FP8 sibling together with the hosted qwen3.8-27b API release of August 2026. The 2.4T Max flagship and its custom license, the qwen3.8-max-0902 snapshot, the max-preview and flash API lines, the experimental Flash-Next preview, earlier Qwen generations, QwenCloud product packaging, ModelScope mirrors, and community quantizations remain separate boundaries.", "strengths": [ "Permissively licensed dense 27B checkpoint with verbatim Apache-2.0 license file, documented SGLang, vLLM, and TokenSpeed recipes, and an official FP8 variant", "First-party QwenCloud and Alibaba Cloud Model Studio serving under the exact qwen3.8-27b model ID plus documented Novita third-party serverless hosting" ], "tradeoffs": [ "The downloaded checkpoint is narrower than its API namesake, with smaller native context extended by YaRN versus the API default", "Dense 27B still needs a high-memory single GPU or quantization for laptop-class self-operation" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.42 }, "evidence": [ { "kind": "web", "label": "Official Qwen3.8-27B weights repository and model card", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Apache-2.0 license blob", "url": "https://huggingface.co/Qwen/Qwen3.8-27B/blob/main/LICENSE", "verified_at": "2026-09-12" }, { "kind": "web", "label": "QwenCloud qwen3.8-27b model page", "url": "https://www.qwencloud.com/models/qwen3.8-27b", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Alibaba Cloud Model Studio qwen3.8-27b documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-8-27b", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Novita serverless docs for qwen/qwen3.8-27b", "url": "https://novita.ai/models/model-detail/qwen-qwen3.8-27b", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/alibaba/qwen3.8-27b.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12" }, { "id": "model-alibaba-qwen3-8-flash", "source_id": "alibaba/qwen3.8-flash", "name": "Qwen3.8 Flash", "developer": "Alibaba", "url": "https://help.aliyun.com/en/model-studio/qwen3-8-flash", "description": "Alibaba's latest multimodal Qwen release for coding assistance and agentic workflows, served as the qwen3.8-flash API model with 1M context and no downloadable weights under its own name.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Qwen3.8 Flash", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "family": "qwen", "release_date": "2026-08-26", "last_updated": "2026-08-26", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Alibaba-Model-Studio-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Qwen3.8-Flash is served as a Model Studio API model governed by the Alibaba Cloud Model Studio Service Agreement (effective September 2, 2026) and its service-specific terms. Those terms govern managed API access with fees, reserve all intellectual property to Alibaba, prohibit reverse engineering and resale of the service or its models, and bar training competing products on outputs; they do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Alibaba-Model-Studio-Agreement", "scope": "Business and developer access to the qwen3.8-flash model through Alibaba Cloud Model Studio API services", "kind": "web_terms", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the qwen3.8-flash API line served by Alibaba Cloud Model Studio under invocation ID qwen3.8-flash, with text, image, and video input and 1M context. The Qwen3.8-Flash-Next preview weights, the qwen3.8-max line, earlier Qwen generations, QwenCloud product packaging, and community quantizations remain separate boundaries.", "strengths": [ "Alibaba documents a single invocation ID with text, image, and video input, function calling, and 1M context across six serving regions", "OpenAI- and Anthropic-compatible API protocols are documented for the line" ], "tradeoffs": [ "No downloadable weights exist under the qwen3.8-flash name; self-operation is not available for this line", "Access, quotas, and lifecycle remain controlled by the managed API rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official qwen3.8-flash model, modality, and endpoint documentation", "url": "https://help.aliyun.com/en/model-studio/qwen3-8-flash", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Alibaba Cloud Model Studio Service Agreement", "url": "https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Model Studio related agreements: governing terms index", "url": "https://help.aliyun.com/en/model-studio/related-agreements", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwen3.8-flash.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwen3-8-flash-next", "source_id": "alibaba/qwen3.8-flash-next", "name": "Qwen3.8 Flash Next", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "description": "Alibaba's dated experimental preview of the Qwen4 architecture: a hybrid-attention MoE with 6B active parameters per token plus offloadable n-gram embeddings, published as open weights for coding and agentic work while the production qwen3.8-flash API line stays a separate model.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen3.8 Flash Next", "description": "Open-weight experimental preview of the Qwen4 architecture: hybrid-attention MoE (125B total, 6B active) with vision encoder for coding, agent tasks, and image and video understanding", "family": "qwen", "release_date": "2026-08-27", "last_updated": "2026-08-27", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": "qwen-community-1.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next" } ] }, "licenses": [ "LicenseRef-Qwen-Community-1.0" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The downloadable weights carry the Qwen Community License 1.0, which grants use, modification, distribution, hosting, and fine-tuning with attribution above 100 million monthly active users or 20 million US dollars in monthly revenue, but requires any model-as-a-service or AI-work-assistant business to obtain a separate license before any commercial use, with no revenue threshold. That stricter commercial gate distinguishes it from the Max sibling license, so the release is source-available rather than open source. The production qwen3.8-flash API service is governed separately.", "license_evidence": [ { "license_id": "LicenseRef-Qwen-Community-1.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.8-Flash-Next repository; does not cover the hosted qwen3.8-flash API service", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/main/LICENSE", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the fixed, dated Qwen3.8-Flash-Next preview checkpoint released August 26, 2026 as an early look at the Qwen4 architecture — a named reviewable artifact, not a floating tag. The production qwen3.8-flash API model, the Max flagship and its snapshots, the Apache-2.0 Qwen3.8-27B sibling, earlier Qwen generations, Qwen4 proper when it lands, and community quantizations remain separate boundaries.", "strengths": [ "First open-weight release of the Qwen4-track architecture with full model card, tech report, and SGLang, vLLM, and TokenSpeed recipes", "Roughly one-ninth the training cost of the prior generation flagship at competitive coding and agent capability with 6B active parameters per token" ], "tradeoffs": [ "The Community 1.0 weights license forces any model-as-a-service or AI-assistant business into a separate license with no revenue floor", "No inference provider or first-party API serves this exact checkpoint, so self-operation is the only path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 8, "deployment_portability": 4, "serving_reach": 3, "lifecycle_transparency": 5, "documentation_provenance": 8, "overall": 5.26 }, "evidence": [ { "kind": "web", "label": "Official Qwen3.8-Flash-Next weights repository and model card", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Qwen Community License 1.0 blob", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/main/LICENSE", "verified_at": "2026-09-12" }, { "kind": "web", "label": "QwenLM release repository with August 2026 changelog entry", "url": "https://github.com/QwenLM/Qwen3.8-Flash-Next/", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Release report: production version served as qwen3.8-flash on QwenCloud", "url": "https://www.alibabacloud.com/blog/qwen-3-8-flash-next-a-new-architecture-towards-ultimate-cost-efficiency_603501", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/alibaba/qwen3.8-flash-next.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12" }, { "id": "model-alibaba-qwen3-8-max", "source_id": "alibaba/qwen3.8-max", "name": "Qwen3.8 Max", "developer": "Alibaba", "url": "https://www.qwencloud.com/models/qwen3.8-max", "description": "Alibaba's 2.4-trillion-parameter MoE flagship for coding, professional work, and long-horizon agentic workflows, served as the multimodal qwen3.8-max API model with 1M context and published as text-only open weights under a custom license.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Qwen3.8 Max", "description": "2.4-trillion-parameter MoE flagship for coding, professional work, multimodal understanding, and long-horizon agentic workflows", "family": "qwen", "release_date": "2026-08-03", "last_updated": "2026-08-03", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Qwen3.8-Max" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The downloadable weights carry the custom Qwen3.8-Max License, which grants use, modification, distribution, hosting, and fine-tuning but requires attribution above 100 million monthly active users or 20 million US dollars in monthly revenue and a separate license for model-as-a-service or AI-work-assistant businesses above 50 million US dollars in trailing revenue. Hosted qwen3.8-max API use is governed separately by QwenCloud and Alibaba Cloud Model Studio service terms, not by the weights license.", "license_evidence": [ { "license_id": "LicenseRef-Qwen3.8-Max", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/Qwen3.8-2.4T-A95B Hugging Face repository; does not cover the hosted qwen3.8-max API service", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/blob/main/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the qwen3.8-max line, now serving the September 2, 2026 snapshot (qwen3.8-max-0902, alias qwen3.8-max-2026-09-02) by default since the September 5 upgrade, together with its open-weights checkpoint Qwen/Qwen3.8-2.4T-A95B and the official FP8 sibling. The downloaded checkpoint is the August weights release and is narrower than the API namesake (text-only, thinking-required, 262K native context versus the API's vision input and 1M default context). The qwen3.8-max-preview API preview, the Apache-2.0 Qwen3.8-27B sibling, earlier Qwen generations, QwenCloud product packaging, and community quantizations remain separate boundaries.", "strengths": [ "The first Max-class Qwen with downloadable weights (2.4T total, 95B active MoE) plus documented vLLM, SGLang, and TokenSpeed recipes and an official FP8 variant", "First-party QwenCloud and Alibaba Cloud Model Studio serving with 1M context, both exposing the pinnable qwen3.8-max-0902 snapshot, plus a documented NVIDIA NIM third-party path for the exact checkpoint" ], "tradeoffs": [ "The custom weights license gates commercial model-as-a-service use above a revenue threshold behind a separate license, unlike earlier Apache-2.0 Qwen generations", "The 4.89 TB BF16 checkpoint needs multi-node serving, putting self-operation out of reach for most operators" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 4, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 6.68 }, "evidence": [ { "kind": "web", "label": "QwenCloud qwen3.8-max model page", "url": "https://www.qwencloud.com/models/qwen3.8-max", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Alibaba Cloud Model Studio qwen3.8-max documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-8-max", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Qwen3.8-2.4T-A95B weights repository and model card", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "verified_at": "2026-09-11" }, { "kind": "web", "label": "NVIDIA NIM guide for the Qwen3.8-2.4T-A95B checkpoint", "url": "https://docs.nvidia.com/nim/large-language-models/2.0.10/get-started/advanced/get-started-qwen3.8.html", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/alibaba/qwen3.8-max.toml", "verified_at": "2026-09-11" }, { "kind": "web", "label": "QwenCloud qwen3.8-max-0902 snapshot page (default serving snapshot since September 5, 2026)", "url": "https://www.qwencloud.com/models/qwen3.8-max-0902", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-12" }, { "id": "model-alibaba-qwen3-8-omni-flash", "source_id": "alibaba/qwen3.8-omni-flash", "name": "Qwen3.8 Omni Flash", "developer": "Alibaba", "url": "https://help.aliyun.com/en/model-studio/qwen3-8-omni-flash", "description": "Alibaba's Qwen3.8 Omni Flash, an omni-modal model accepting text, image, audio, and video input with 1M context and enabled thinking, served through Alibaba Cloud Model Studio across six regions and QwenCloud.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Qwen3.8 Omni Flash", "description": "Qwen omni model for text, vision, audio, and multimodal agent tasks", "family": "qwen", "release_date": "2026-09-17", "last_updated": "2026-09-17", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Alibaba-Model-Studio-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Qwen3.8 Omni Flash is served, not distributed, under the Alibaba Cloud Model Studio service agreement that governs the other reviewed hosted Qwen records.", "license_evidence": [ { "license_id": "LicenseRef-Alibaba-Model-Studio-Agreement", "scope": "Qwen3.8 Omni Flash hosted model and API served under Alibaba Cloud Model Studio terms", "kind": "web_terms", "url": "https://help.aliyun.com/en/model-studio/qwen3-8-omni-flash", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Qwen3.8 Omni Flash hosted release served through Alibaba Cloud Model Studio, the successor to the qwen3.5-omni line. The qwen3.8-flash, qwen3.8-max, and open-weight Qwen releases are separate boundaries.", "strengths": [ "Distinct new omni-modal line with fully documented native audio and video input and 1M context", "Dated docs and a live serving presence across six regions and QwenCloud" ], "tradeoffs": [ "No downloadable weights, so self-operation is unavailable", "A very fresh release with the lifecycle still settling" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/alibaba/qwen3.8-omni-flash.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://help.aliyun.com/en/model-studio/qwen3-8-omni-flash", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-alibaba-qwq-32b", "source_id": "alibaba/qwq-32b", "name": "QwQ 32B", "developer": "Alibaba", "url": "https://huggingface.co/Qwen/QwQ-32B", "description": "Alibaba's 32-billion-parameter QwQ reasoning release, published as Apache-2.0 open weights with hosted API access.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "QwQ 32B", "description": "Open reasoning model from the Qwen team for math, coding, and step-by-step problem solving", "family": "qwen", "release_date": "2025-03-05", "last_updated": "2025-03-05", "knowledge_cutoff": "2024-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/QwQ-32B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, appendix completed as Copyright 2024 Alibaba Cloud, permitting use, modification, distribution, hosting, and fine-tuning with no attribution gate, revenue threshold, or separate model-as-a-service license. Hosted API use is governed separately by Alibaba Cloud Model Studio service terms.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Model weights, parameters, configuration files, inference code, and associated documentation in the Qwen/QwQ-32B repository", "kind": "web_terms", "url": "https://huggingface.co/Qwen/QwQ-32B/blob/976055f8c83f394f35dbd3ab09a285a984907bd0/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the Qwen/QwQ-32B weight release of March 2025 at Hugging Face commit 976055f8c83f394f35dbd3ab09a285a984907bd0 (last modified 2026-03-11), together with hosted qwq-32b API access. The QwQ-32B-Preview, later Qwen reasoning lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 32B reasoning checkpoint carrying the verbatim Apache-2.0 license file beside the weights", "Compact size keeps full-precision self-operation within reach of smaller GPU setups" ], "tradeoffs": [ "No current exact-ID Model Studio model page documents the hosted endpoint", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.82 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/Qwen/QwQ-32B", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official QwQ-32B first-party release blog", "url": "https://qwenlm.github.io/blog/qwq-32b/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwq-32b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-alibaba-qwq-plus", "source_id": "alibaba/qwq-plus", "name": "QwQ-Plus", "developer": "Alibaba", "url": "https://www.alibabacloud.com/help/en/model-studio/qwq-plus", "description": "Alibaba's hosted QwQ-Plus reasoning model, an enhanced counterpart to the QwQ line trained on the Qwen2.5 base with reinforcement learning, served through Alibaba Cloud Model Studio.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "QwQ Plus", "description": "Qwen reasoning model for deliberate problem solving, math, and coding", "family": "qwen", "release_date": "2025-03-05", "last_updated": "2025-03-05", "knowledge_cutoff": "2024-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Alibaba-Model-Studio-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The hosted QwQ-Plus model is governed by Alibaba Cloud Model Studio service terms; the Qwen/QwQ-Plus repository is gated and no weights are distributed.", "license_evidence": [ { "license_id": "LicenseRef-Alibaba-Model-Studio-Agreement", "scope": "QwQ-Plus hosted model and API served under Alibaba Cloud Model Studio terms", "kind": "web_terms", "url": "https://www.alibabacloud.com/help/en/model-studio/qwq-plus", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the QwQ-Plus hosted reasoning model served through Alibaba Cloud Model Studio with the qwq-plus API identifier. The open-weight QwQ checkpoints, the qwen-plus, qwen-max, and Qwen3 families, and independent hosts are separate boundaries.", "strengths": [ "Dated first-party Model Studio page establishes the hosted identity and API identifier", "Explicit reasoning_content thinking-token output is documented alongside the API" ], "tradeoffs": [ "Closed weights with a gated repository and no accessible checkpoint", "Built on the older Qwen2.5 base rather than a current-generation foundation" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 4.14 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/alibaba/qwq-plus.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://www.alibabacloud.com/help/en/model-studio/qwq-plus", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-amazon-nova-2-lite", "source_id": "amazon/nova-2-lite", "name": "Nova 2 Lite", "developer": "Amazon", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-lite.html", "description": "An active multimodal Nova release served through Amazon Bedrock under the AWS Service Terms, with a 1,000,000-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nova 2 Lite", "description": "Multimodal reasoning model for visual analysis, planning, and tool use", "family": "nova", "release_date": "2025-12-02", "last_updated": "2025-12-01", "knowledge_cutoff": "2025-10", "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-AWS-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "AWS provides the release as a managed service under the AWS Customer Agreement and the AWS Service Terms Section 60.3 for Amazon Bedrock, which cover the Amazon Nova and Titan foundation models and grant exclusive use of customized Nova models. The terms permit customer applications and Bedrock and SageMaker customization but license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-AWS-Service-Terms", "kind": "web_terms", "url": "https://aws.amazon.com/service-terms/", "verified_at": "2026-09-18", "scope": "Access to the release through Amazon Bedrock under the AWS Service Terms Section 60.3 and the AWS Customer Agreement" } ], "access_boundary": "This record represents the Amazon Nova 2 Lite release identified by the Bedrock model ID amazon.nova-2-lite-v1:0, launched December 2, 2025. The Nova 1 line, the other Nova 2 sizes, the Sonic, Canvas, and Reel lines, SageMaker and the other Bedrock surfaces, and later Nova releases remain separate service or product boundaries.", "strengths": [ "AWS publishes a model card with launch date, lifecycle status, context limits, and the Bedrock model ID", "The model card documents the global Bedrock region reach and geo-specific model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through Amazon Bedrock with no other documented host path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official Amazon Nova 2 Lite model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-lite.html", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Nova 2 launch announcement", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-lite-a-fast-cost-effective-reasoning-model/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/amazon/nova-2-lite.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-amazon-nova-lite", "source_id": "amazon/nova-lite", "name": "Nova Lite", "developer": "Amazon", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-lite.html", "description": "An active multimodal Nova release served through Amazon Bedrock under the AWS Service Terms, with a 300,000-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nova Lite", "description": "Efficient model for low-latency assistance, extraction, and routine automation", "family": "nova-lite", "release_date": "2024-12-03", "last_updated": "2024-12-03", "knowledge_cutoff": "2024-10", "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 300000, "input": null, "output": 10000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-AWS-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "AWS provides the release as a managed service under the AWS Customer Agreement and the AWS Service Terms Section 60.3 for Amazon Bedrock, which cover the Amazon Nova and Titan foundation models and grant exclusive use of customized Nova models. The terms permit customer applications and Bedrock and SageMaker customization but license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-AWS-Service-Terms", "kind": "web_terms", "url": "https://aws.amazon.com/service-terms/", "verified_at": "2026-09-18", "scope": "Access to the release through Amazon Bedrock under the AWS Service Terms Section 60.3 and the AWS Customer Agreement" } ], "access_boundary": "This record represents the Amazon Nova Lite release identified by the Bedrock model ID amazon.nova-lite-v1:0, launched December 5, 2024. The Nova 2 Lite successor, the Micro and Pro siblings, the Sonic, Canvas, and Reel lines, SageMaker and the other Bedrock surfaces, and later Nova releases remain separate service or product boundaries.", "strengths": [ "AWS publishes a model card with launch date, lifecycle status, context limits, and the Bedrock model ID", "The model card documents the global Bedrock region reach and geo-specific model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through Amazon Bedrock with no other documented host path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Official Amazon Nova Lite model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-lite.html", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Nova launch announcement", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/amazon/nova-lite.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-amazon-nova-micro", "source_id": "amazon/nova-micro", "name": "Nova Micro", "developer": "Amazon", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-micro.html", "description": "An active text-only Nova release served through Amazon Bedrock under the AWS Service Terms, with a 128,000-token context and no downloadable model weights.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nova Micro", "description": "Efficient model for low-latency assistance, extraction, and routine automation", "family": "nova-micro", "release_date": "2024-12-03", "last_updated": "2024-12-03", "knowledge_cutoff": "2024-10", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 10000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-AWS-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "AWS provides the release as a managed service under the AWS Customer Agreement and the AWS Service Terms Section 60.3 for Amazon Bedrock, which cover the Amazon Nova and Titan foundation models and grant exclusive use of customized Nova models. The terms permit customer applications and Bedrock and SageMaker customization but license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-AWS-Service-Terms", "kind": "web_terms", "url": "https://aws.amazon.com/service-terms/", "verified_at": "2026-09-18", "scope": "Access to the release through Amazon Bedrock under the AWS Service Terms Section 60.3 and the AWS Customer Agreement" } ], "access_boundary": "This record represents the Amazon Nova Micro release identified by the Bedrock model ID amazon.nova-micro-v1:0, launched December 5, 2024. The Lite and Pro siblings, the Nova 2 generation, the Sonic, Canvas, and Reel lines, SageMaker and the other Bedrock surfaces, and later Nova releases remain separate service or product boundaries.", "strengths": [ "AWS publishes a model card with launch date, lifecycle status, context limits, and the Bedrock model ID", "The model card documents the global Bedrock region reach and geo-specific model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through Amazon Bedrock with no other documented host path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Official Amazon Nova Micro model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-micro.html", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Nova launch announcement", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/amazon/nova-micro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-amazon-nova-pro", "source_id": "amazon/nova-pro", "name": "Nova Pro", "developer": "Amazon", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-pro.html", "description": "An active multimodal Nova release served through Amazon Bedrock under the AWS Service Terms, with a 300,000-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nova Pro", "description": "Flagship model for demanding analysis, coding, and production agent workflows", "family": "nova-pro", "release_date": "2024-12-03", "last_updated": "2024-12-03", "knowledge_cutoff": "2024-10", "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 300000, "input": null, "output": 10000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-AWS-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "AWS provides the release as a managed service under the AWS Customer Agreement and the AWS Service Terms Section 60.3 for Amazon Bedrock, which cover the Amazon Nova and Titan foundation models and grant exclusive use of customized Nova models. The terms permit customer applications and Bedrock and SageMaker customization but license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-AWS-Service-Terms", "kind": "web_terms", "url": "https://aws.amazon.com/service-terms/", "verified_at": "2026-09-18", "scope": "Access to the release through Amazon Bedrock under the AWS Service Terms Section 60.3 and the AWS Customer Agreement" } ], "access_boundary": "This record represents the Amazon Nova Pro release identified by the Bedrock model ID amazon.nova-pro-v1:0, launched December 5, 2024. The Lite and Micro siblings, the Nova 2 Lite migration target, the Sonic, Canvas, and Reel lines, SageMaker and the other Bedrock surfaces, and later Nova releases remain separate service or product boundaries.", "strengths": [ "AWS publishes a model card with launch date, lifecycle status, context limits, and the Bedrock model ID", "The model card documents the global Bedrock region reach and geo-specific model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through Amazon Bedrock with no other documented host path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Official Amazon Nova Pro model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-pro.html", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Nova launch announcement", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/amazon/nova-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-anthropic-claude-fable-5", "source_id": "anthropic/claude-fable-5", "name": "Claude Fable 5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/fable-5/overview", "description": "A pinned Mythos-class Claude release in legacy status with a 1M-token context, available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Fable 5", "description": "Claude model for creative writing, analysis, and controlled agent workflows", "family": "claude-fable", "release_date": "2026-06-09", "last_updated": "2026-06-09", "knowledge_cutoff": "2026-01-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides the release as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-18", "scope": "Commercial access to the release through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the Claude Fable 5 release identified by the dateless pinned ID claude-fable-5, the general-use safeguarded variant of the Mythos-class tier that sits above the Opus class. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, Claude Mythos 5, and later Fable releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed dateless model ID with pinned-snapshot semantics, legacy status, and a retirement date commitment", "The suspension and redeployment events of June and July 2026 are documented first-party" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is in legacy status with a retirement commitment only until June 9, 2027" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Fable 5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/fable-5/overview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude Fable 5 launch announcement", "url": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/anthropic/claude-fable-5.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "ASL-3", "domains": [ "bio_chem" ], "determination": "safeguard_standard", "scope": "deployment", "statement": "As with many previous models, we are treating Mythos 5 as CB-1 and applying ASL-3 protections to Fable 5, including blocking classifiers against CB-1 misuse.", "url": "https://www.anthropic.com/claude-fable-5-mythos-5-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-fable-5-1", "source_id": "anthropic/claude-fable-5-1", "name": "Claude Fable 5.1", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/fable-5-1/overview", "description": "A pinned Mythos-class Claude release in current status with a 1M-token context, available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Fable 5.1", "description": "Claude model for demanding reasoning and long-horizon agentic work", "family": "claude-fable", "release_date": "2026-09-01", "last_updated": "2026-09-01", "knowledge_cutoff": "2026-06", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides the release as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-18", "scope": "Commercial access to the release through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the Claude Fable 5.1 release identified by the dateless pinned ID claude-fable-5-1, the current model of the Fable line with a documented breaking-change surface against Fable 5. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, Claude Fable 5, and later releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed model ID with pinned-snapshot semantics, lifecycle status, release date, and a retirement date commitment", "The official model page identifies first-party API access and independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Three documented API changes are breaking for developers migrating from Fable 5" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Fable 5.1 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/fable-5-1/overview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude Fable 5.1 release announcement", "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/anthropic/claude-fable-5-1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "deployment", "statement": "On chemical and biological risks, we judge that the model has CB-1 capabilities—meaning it could meaningfully help someone with a basic technical background synthesize a known weapon—but falls short of the CB-2 threshold for functionally replacing rare expert talent, which is the limiting factor in developing novel weapons. We hold this judgment with some uncertainty, and are deploying Claude Fable 5.1 with the same biological safeguards we deployed with Claude Fable 5.", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "low" } ] }, { "id": "model-anthropic-claude-haiku-4-5", "source_id": "anthropic/claude-haiku-4-5", "name": "Claude Haiku 4.5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/haiku-4-5/overview", "description": "A dated, pinned multimodal Claude release available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Haiku 4.5 (latest)", "description": "Fast Claude lane for lightweight agents, office tasks, and responsive chat", "family": "claude-haiku", "release_date": "2025-10-15", "last_updated": "2025-10-15", "knowledge_cutoff": "2025-02-28", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 64000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides Claude Haiku 4.5 as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-11", "scope": "Commercial access to Claude Haiku 4.5 through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the fixed Claude Haiku 4.5 snapshot identified by claude-haiku-4-5-20251001, which the claude-haiku-4-5 alias resolves to. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and any later Haiku release remain separate service or product boundaries.", "strengths": [ "Anthropic documents a dated snapshot ID, alias behaviour, active lifecycle status, release date, and a retirement date commitment", "The official model page identifies first-party API access and three independent cloud-platform paths, including both Bedrock integrations" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access and continued availability depend on managed-service terms, regional eligibility, and provider lifecycle decisions, with partner clouds setting their own retirement dates" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Haiku 4.5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/haiku-4-5/overview", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/anthropic/claude-haiku-4-5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "ASL-2", "domains": [ "bio_chem" ], "determination": "safeguard_standard", "scope": "weights", "statement": "Our safety testing also showed that Claude Haiku 4.5 poses only limited risks in terms of the production of chemical, biological, radiological, and nuclear (CBRN) weapons. For that reason, we’ve released it under the AI Safety Level 2 (ASL-2) standard—compared to the more restrictive ASL-3 for Sonnet 4.5 and Opus 4.1.", "url": "https://www.anthropic.com/news/claude-haiku-4-5", "content_sha256": "a0c708ac879bc9a51a52d72cb37a310ffbbf582ec33c4074b5ca363028e2f139", "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-mythos-5", "source_id": "anthropic/claude-mythos-5", "name": "Claude Mythos 5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/mythos-5/overview", "description": "An invite-only pinned Mythos-class Claude release with a 1M-token context, available through Anthropic's developer API under access programs and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Mythos 5", "description": "Restricted Claude model for advanced cybersecurity and biology research workflows", "family": "claude-mythos", "release_date": "2026-06-09", "last_updated": "2026-06-09", "knowledge_cutoff": "2026-01-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides the release as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-18", "scope": "Commercial access to the release through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the Claude Mythos 5 release identified by the dateless pinned ID claude-mythos-5, the safeguards-lifted variant of the Mythos-class tier delivered by invitation through Project Glasswing and sharing Claude Fable 5's underlying weights. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Fable 5, Claude Mythos Preview, and later Mythos releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed dateless model ID with pinned-snapshot semantics, invite-only status, and a retirement date commitment", "The suspension and redeployment events of June and July 2026 are documented first-party" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access is invite-only through the trusted-access programs, gating reach to approved customers" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official Claude Mythos 5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/mythos-5/overview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude Fable 5 and Mythos 5 launch announcement", "url": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/anthropic/claude-mythos-5.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "We therefore treat Claude Mythos 5 as having CB-1 capabilities and apply commensurate protections: real-time classifier guards, access controls for guard exemptions, a bug bounty program and threat intelligence, rapid-response options for jailbreaks, and security controls against model weight theft.", "url": "https://www.anthropic.com/claude-fable-5-mythos-5-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-mythos-5-1", "source_id": null, "name": "Claude Mythos 5.1", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/mythos-5-1/overview", "description": "An invite-only pinned Mythos-class Claude release with a 1M-token context, available through Anthropic's developer API under Project Glasswing and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Mythos 5.1", "description": null, "family": null, "release_date": "2026-09-01", "last_updated": null, "knowledge_cutoff": "2026-06", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": null, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides the release as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-20", "scope": "Commercial access to the release through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the Claude Mythos 5.1 release identified by the dateless pinned ID claude-mythos-5-1, which Anthropic documents as the same model as Claude Fable 5.1 offered by invitation only through Project Glasswing. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Fable 5.1, Claude Mythos 5, and later Mythos releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed dateless model ID with pinned-snapshot semantics, invite-only status, a release date, and a retirement date commitment", "The official model page lists per-platform model IDs for the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access is invite-only through Project Glasswing and an account team, gating reach to approved customers" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official Claude Mythos 5.1 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/mythos-5-1/overview", "verified_at": "2026-09-20" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-20" }, { "kind": "web", "label": "Official Claude Fable 5.1 and Mythos 5.1 release announcement", "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1", "verified_at": "2026-09-20" } ], "metadata_verified_at": "2026-09-20", "verified_at": "2026-09-20", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "We apply the same CB-1 designation and safeguards to Mythos 5.1 as we applied to Mythos 5. We emphasize that although we have not applied the CB-2 designation to either Mythos 5 or Mythos 5.1, we have conservatively expanded the safeguards applied to these models relative to those we applied to weaker models, including Opus 5.", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-opus-4-5-20251101", "source_id": "anthropic/claude-opus-4-5-20251101", "name": "Claude Opus 4.5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/opus-4-5/overview", "description": "A pinned multimodal Claude release in legacy status, available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Opus 4.5", "description": "Flagship Claude model for deep reasoning, coding, and long-horizon agents", "family": "claude-opus", "release_date": "2025-11-01", "last_updated": "2025-11-01", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 64000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides the release as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-18", "scope": "Commercial access to the release through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the fixed Claude Opus 4.5 snapshot identified by claude-opus-4-5-20251101, which the pre-4.6 alias claude-opus-4-5 resolves to. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and later Opus releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed model ID with pinned-snapshot semantics, lifecycle status, release date, and a retirement date commitment", "The official model page identifies first-party API access and independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is in legacy status with a retirement commitment only until November 24, 2026" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Opus 4.5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/opus-4-5/overview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude Opus 4.5 release announcement", "url": "https://www.anthropic.com/news/claude-opus-4-5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/anthropic/claude-opus-4-5-20251101.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "ASL-3", "domains": [ "bio_chem" ], "determination": "safeguard_standard", "scope": "weights", "statement": "In particular, Claude Opus 4.5 outperformed Claude Opus 4.1 and Claude Sonnet 4.5 models on both long-form virology tasks, as well as on several subtasks of the DNA synthesis screening evaluation, while achieving comparable performance on all other evaluations. As a result, we determined ASL-3 safeguards were appropriate.", "url": "https://www.anthropic.com/claude-opus-4-5-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-opus-4-6", "source_id": "anthropic/claude-opus-4-6", "name": "Claude Opus 4.6", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/opus-4-6/overview", "description": "A pinned multimodal Claude release in legacy status with a 1M-token context, available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Opus 4.6", "description": "High-end Claude for difficult coding, planning, and slower expert reasoning", "family": "claude-opus", "release_date": "2026-02-05", "last_updated": "2026-03-13", "knowledge_cutoff": "2025-05-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides the release as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-18", "scope": "Commercial access to the release through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the Claude Opus 4.6 release identified by the dateless pinned ID claude-opus-4-6, which Anthropic documents as a single fixed snapshot rather than an alias. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and other Opus releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed model ID with pinned-snapshot semantics, lifecycle status, release date, and a retirement date commitment", "The official model page identifies first-party API access and independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is in legacy status with a retirement commitment only until February 5, 2027" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Opus 4.6 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/opus-4-6/overview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude Opus 4.6 release announcement", "url": "https://www.anthropic.com/news/claude-opus-4-6", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/anthropic/claude-opus-4-6.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "AI R&D-4", "domains": [ "autonomy" ], "determination": "precautionary", "scope": "weights", "statement": "As a result, while we do not believe Claude Opus 4.6 meets the threshold for ASL-4 autonomy safeguards, we find ourselves in a gray zone where clean rule-out is difficult and the margin to the threshold is unclear. We expect with high probability that models in the near future could cross this threshold. Given this uncertainty, we have taken a cautious approach, similarly to how we approached the ASL-3 CBRN threshold for Claude Opus 4, and have chosen to prepare and publish a Sabotage Risk Report for Claude Opus 4.6, consistent with the RSP’s commitment to developing an affirmative case addressing misalignment risks for the AI R&D-4 threshold.", "url": "https://www.anthropic.com/claude-opus-4-6-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "medium" } ] }, { "id": "model-anthropic-claude-opus-4-7", "source_id": "anthropic/claude-opus-4-7", "name": "Claude Opus 4.7", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/opus-4-7/overview", "description": "A pinned multimodal Claude release in legacy status with a 1M-token context and an updated tokenizer, available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Opus 4.7", "description": "Stronger Opus tier for advanced software work and high-stakes reasoning", "family": "claude-opus", "release_date": "2026-04-16", "last_updated": "2026-04-16", "knowledge_cutoff": "2026-01-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides the release as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-18", "scope": "Commercial access to the release through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the Claude Opus 4.7 release identified by the dateless pinned ID claude-opus-4-7, which Anthropic documents as a single fixed snapshot rather than an alias. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and other Opus releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed model ID with pinned-snapshot semantics, lifecycle status, release date, and a retirement date commitment", "The official model page identifies first-party API access and independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is in legacy status with a retirement commitment only until April 16, 2027" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Opus 4.7 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/opus-4-7/overview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Claude Opus 4.7 release announcement", "url": "https://www.anthropic.com/news/claude-opus-4-7", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/anthropic/claude-opus-4-7.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "Autonomy threat model 1", "domains": [ "autonomy" ], "determination": "determined", "scope": "weights", "statement": "Autonomy threat model 1 is applicable to Claude Opus 4.7, as it is to some of our previous AI models.", "url": "https://www.anthropic.com/claude-opus-4-7-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "medium" } ] }, { "id": "model-anthropic-claude-opus-4-8", "source_id": "anthropic/claude-opus-4-8", "name": "Claude Opus 4.8", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/opus-4-8/overview", "description": "A pinned multimodal Claude release in legacy status, available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Opus 4.8", "description": "Top Claude Opus tier for the hardest reasoning, coding, and long-horizon agents", "family": "claude-opus", "release_date": "2026-05-28", "last_updated": "2026-05-28", "knowledge_cutoff": "2026-01", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides Claude Opus 4.8 as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-11", "scope": "Commercial access to Claude Opus 4.8 through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the fixed Claude Opus 4.8 model release identified by the dateless pinned ID claude-opus-4-8, which Anthropic documents as a single fixed snapshot rather than an alias. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and Claude Opus 5 and later Opus releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed dateless model ID with pinned-snapshot semantics, legacy lifecycle status, release date, and a retirement date commitment", "The official model page identifies first-party API access and three independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is in legacy status with a retirement commitment only until May 28, 2027, and partner clouds set their own retirement dates" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Opus 4.8 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/opus-4-8/overview", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/anthropic/claude-opus-4-8.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Non-novel chemical and biological weapons (CB-1): We believe it is hard to be confident whether a model passes this threshold, but (as with previous recent models) our results are consistent with Opus 4.8 providing significant uplift to individuals or groups with basic technical backgrounds for this threat model.", "url": "https://www.anthropic.com/claude-opus-4-8-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-opus-5", "source_id": "anthropic/claude-opus-5", "name": "Claude Opus 5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/opus-5/overview", "description": "A pinned multimodal Claude Opus release available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Opus 5", "description": "Strongest Claude Opus model for coding, agents, and professional work", "family": "claude-opus", "release_date": "2026-07-24", "last_updated": "2026-07-24", "knowledge_cutoff": "2026-05", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides Claude Opus 5 as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "scope": "Commercial access to Claude Opus 5 through Anthropic API keys and related Anthropic services, including incorporated usage restrictions", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the fixed Claude Opus 5 release identified by the dateless pinned ID claude-opus-5, which Anthropic documents as a single fixed snapshot rather than an alias. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and other Claude releases including Opus 4.8 remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed dateless model ID with pinned-snapshot semantics, active lifecycle status, a July 24, 2026 release date, and a retirement commitment of not sooner than July 24, 2027", "The official model page identifies first-party API access and three independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access and continued availability depend on managed-service terms, regional eligibility, and provider lifecycle decisions" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Opus 5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/opus-5/overview", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Anthropic Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-11" }, { "kind": "web", "label": "AWS Bedrock model card showing Claude Opus 5 availability", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-opus-5.html", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Google Cloud model card showing Claude Opus 5 availability", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/claude/opus-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Microsoft Foundry model catalog showing claude-opus-5 availability", "url": "https://ai.azure.com/catalog/models/claude-opus-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/anthropic/claude-opus-5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "It is difficult to say with full confidence whether a model passes this threshold. However, Claude Opus 5 is broadly more capable than previous models like Sonnet 4.6 which we have conservatively treated as having CB-1 capabilities. As with earlier models we have conservatively treated as having CB-1 capabilities, we apply commensurate protections: real-time classifier guards, access controls for guard exemptions, a bug bounty program and threat intelligence, rapid-response options for jailbreaks, and security controls against model weight theft.", "url": "https://www.anthropic.com/claude-opus-5-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "medium" } ] }, { "id": "model-anthropic-claude-opus-5-5", "source_id": "anthropic/claude-opus-5-5", "name": "Claude Opus 5.5", "developer": "Anthropic", "url": "https://www.anthropic.com/claude-opus-5-5", "description": "Anthropic's Claude Opus 5.5, the first model of the Claude 5.5 family released September 22, 2026, a text-and-image model with 1M-token context and always-on adaptive thinking, served through the Claude API and major cloud marketplaces.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Opus 5.5", "description": "Claude model for long-running agentic coding and knowledge work", "family": "claude-opus", "release_date": "2026-09-22", "last_updated": "2026-09-22", "knowledge_cutoff": "2026-06", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Claude Opus 5.5 is served, not distributed, under Anthropic's commercial terms, which already govern the reviewed Opus and Opus 5 lines. No weights are released.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "scope": "Claude Opus 5.5 API access and related Anthropic services", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Claude Opus 5.5 release of September 22, 2026, the first model of the Claude 5.5 family with the pinned claude-opus-5-5 API identifier. The reviewed Opus 5 and Opus 4.8 lines, earlier Opus releases, and independent hosts are separate boundaries.", "strengths": [ "Dated first-party announcement and docs page with per-platform model identifiers and a retirement commitment of no sooner than September 2027", "Served on the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry" ], "tradeoffs": [ "Closed weights served only through managed APIs", "Always-on thinking cannot be disabled and forced tool-use errors break some Opus 5 migration paths" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.94 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/anthropic/claude-opus-5-5.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://www.anthropic.com/claude-opus-5-5", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "On chemical and biological risks, we treat Opus 5.5 as having CB-1 capabilities (relating to the synthesis of non-novel weapons) but not CB-2 capabilities (relating to the synthesis of novel weapons).", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "medium" } ] }, { "id": "model-anthropic-claude-sonnet-4-5", "source_id": "anthropic/claude-sonnet-4-5", "name": "Claude Sonnet 4.5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/sonnet-4-5/overview", "description": "A dated, pinned multimodal Claude release in legacy status, available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Sonnet 4.5 (latest)", "description": "Balanced Claude model for coding, analysis, agent workflows, and cost control", "family": "claude-sonnet", "release_date": "2025-09-29", "last_updated": "2025-09-29", "knowledge_cutoff": "2025-07-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 64000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides Claude Sonnet 4.5 as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-11", "scope": "Commercial access to Claude Sonnet 4.5 through Anthropic API keys and related Anthropic services, including incorporated usage restrictions" } ], "access_boundary": "This record represents the fixed Claude Sonnet 4.5 snapshot identified by claude-sonnet-4-5-20250929, which the claude-sonnet-4-5 alias resolves to. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and later Sonnet releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a dated snapshot ID, alias behaviour, legacy lifecycle status, release date, and a retirement date commitment", "The official model page identifies first-party API access and three independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is in legacy status with a retirement commitment only until September 29, 2026, and partner clouds set their own retirement dates" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Sonnet 4.5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/sonnet-4-5/overview", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Claude model deprecation policy and lifecycle table", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/anthropic/claude-sonnet-4-5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "ASL-3", "domains": [ "bio_chem" ], "determination": "safeguard_standard", "scope": "weights", "statement": "Claude Sonnet 4.5 is being released under our AI Safety Level 3 (ASL-3) protections, as per our framework that matches model capabilities with appropriate safeguards. These safeguards include filters called classifiers that aim to detect potentially dangerous inputs and outputs—in particular those related to chemical, biological, radiological, and nuclear (CBRN) weapons.", "url": "https://www.anthropic.com/news/claude-sonnet-4-5", "content_sha256": "0228e015950e56918ab2be787f04f55dc0515624f96e38a1b537dd78a71f3397", "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-sonnet-4-6", "source_id": "anthropic/claude-sonnet-4-6", "name": "Claude Sonnet 4.6", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/sonnet-4-6/overview", "description": "A pinned multimodal Claude release available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Sonnet 4.6", "description": "Claude workhorse for coding agents, careful analysis, and production cost control", "family": "claude-sonnet", "release_date": "2026-02-17", "last_updated": "2026-03-13", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 64000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides Claude Sonnet 4.6 as a managed service under its Commercial Terms of Service and incorporated policies. The terms permit customer applications but do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "scope": "Commercial access to Claude Sonnet 4.6 through Anthropic API keys and related Anthropic services, including incorporated usage restrictions", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-04" } ], "access_boundary": "This record represents the fixed Claude Sonnet 4.6 model release identified by claude-sonnet-4-6. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude applications, and later Sonnet releases remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed release ID, lifecycle status, release date, and retirement commitment", "The official model page identifies first-party API access and three independent cloud-platform paths" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access and continued availability depend on managed-service terms, regional eligibility, and provider lifecycle decisions" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Sonnet 4.6 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/sonnet-4-6/overview", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/ef71a4865a5821d8cd24f171a0100156a02e808e/models/anthropic/claude-sonnet-4-6.toml", "verified_at": "2026-09-04" } ], "metadata_verified_at": "2026-09-04", "verified_at": "2026-09-04", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "ASL-3", "domains": [ "bio_chem" ], "determination": "safeguard_standard", "scope": "weights", "statement": "Based on the results above, the Responsible Scaling Officer (RSO) determined the ASL-3 safeguards level to be appropriate for the CBRN domain for Claude Sonnet 4.6.", "url": "https://www.anthropic.com/claude-sonnet-4-6-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-sonnet-5", "source_id": "anthropic/claude-sonnet-5", "name": "Claude Sonnet 5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/sonnet-5/overview", "description": "A pinned multimodal Claude Sonnet release available through Anthropic's developer API and documented third-party cloud platforms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Sonnet 5", "description": "Everyday Claude agent model for coding, planning, browsing, and general work", "family": "claude-sonnet", "release_date": "2026-06-30", "last_updated": "2026-06-30", "knowledge_cutoff": "2026-01-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Anthropic provides Claude Sonnet 5 as a managed service under its Commercial Terms of Service, which incorporate the Usage Policy and Supported Regions Policy by reference. The terms permit customer applications but prohibit reverse engineering the Services or using them to train competing models, and license no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "scope": "Commercial access to Claude Sonnet 5 through Anthropic API keys and related Anthropic services, including incorporated usage restrictions", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the fixed Claude Sonnet 5 release identified by the dateless pinned ID claude-sonnet-5, which Anthropic documents as a single fixed snapshot rather than an alias. Anthropic's API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS, Claude applications, and other Claude releases including Sonnet 4.6 and Opus 5 remain separate service or product boundaries.", "strengths": [ "Anthropic documents a fixed dateless model ID with pinned-snapshot semantics, active lifecycle status, a June 30, 2026 release date, and a retirement commitment of not sooner than June 30, 2027", "The official model page identifies first-party API access and three independent cloud-platform paths with per-platform model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access and continued availability depend on managed-service terms, regional eligibility, and provider lifecycle decisions" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Claude Sonnet 5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/sonnet-5/overview", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Anthropic Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-11" }, { "kind": "web", "label": "AWS Bedrock model card showing Claude Sonnet 5 availability", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-sonnet-5.html", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Google Cloud model card showing Claude Sonnet 5 availability", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/claude/sonnet-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Microsoft Foundry model catalog showing claude-sonnet-5 availability", "url": "https://ai.azure.com/catalog/models/claude-sonnet-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/anthropic/claude-sonnet-5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Non-novel chemical and biological weapons (CB-1): It is hard to be confident about whether a model passes this threshold, but as with previous recent models, our results are consistent with Sonnet 5 providing significant uplift to individuals or groups with basic technical backgrounds for this threat model.", "url": "https://www.anthropic.com/claude-sonnet-5-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "high" } ] }, { "id": "model-anthropic-claude-sonnet-5-5", "source_id": "anthropic/claude-sonnet-5-5", "name": "Claude Sonnet 5.5", "developer": "Anthropic", "url": "https://platform.claude.com/docs/en/models/sonnet-5-5/overview", "description": "The second model of Anthropic's Claude 5.5 family, released September 28, 2026, a text-and-image model with 1M-token context and adaptive thinking, served through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Claude Sonnet 5.5", "description": "Fast Claude model for everyday coding, agents, and knowledge work", "family": "claude-sonnet", "release_date": "2026-09-28", "last_updated": "2026-09-28", "knowledge_cutoff": "2026-06", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Anthropic-Commercial-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Claude Sonnet 5.5 is served, not distributed, under Anthropic's Commercial Terms of Service, which govern the reviewed Sonnet and Opus lines and incorporate the Usage Policy by reference. No weights are released, and the terms prohibit reverse engineering the Services and using them to train competing models.", "license_evidence": [ { "license_id": "LicenseRef-Anthropic-Commercial-Terms", "scope": "Claude Sonnet 5.5 API access and related Anthropic services", "kind": "web_terms", "url": "https://www.anthropic.com/legal/commercial-terms", "verified_at": "2026-09-30" } ], "access_boundary": "This record represents the Claude Sonnet 5.5 release of September 28, 2026, the second model of the Claude 5.5 family with the pinned claude-sonnet-5-5 API identifier and per-platform IDs on Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. The reviewed Sonnet 5, Sonnet 4.6, and Opus 5.5 lines, the unlaunched Claude Haiku 5.5, and independent hosts are separate boundaries.", "strengths": [ "Dated first-party announcement and docs page with per-platform model identifiers, an active latest status, and a retirement commitment of no sooner than September 28, 2027", "Served on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS" ], "tradeoffs": [ "Closed weights served only through managed APIs", "Five documented breaking changes from Sonnet 5, including forced tool use returning an error and up-front thinking requiring the new between_tools setting" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 9, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/fde326351fbf011d3c007163af3046b97efbf6b6/models/anthropic/claude-sonnet-5-5.toml", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Official Claude Sonnet 5.5 model and platform documentation", "url": "https://platform.claude.com/docs/en/models/sonnet-5-5/overview", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Official Claude Sonnet 5.5 release announcement", "url": "https://www.anthropic.com/claude-sonnet-5-5", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Official Claude model ID and versioning policy", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "verified_at": "2026-09-30" }, { "kind": "web", "label": "AWS Bedrock model card showing Claude Sonnet 5.5 availability", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-sonnet-5-5.html", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Google Cloud model card showing Claude Sonnet 5.5 availability", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/claude/sonnet-5-5", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Microsoft Foundry model catalog showing claude-sonnet-5-5 availability", "url": "https://ai.azure.com/catalog/models/claude-sonnet-5-5", "verified_at": "2026-09-30" } ], "metadata_verified_at": "2026-09-30", "verified_at": "2026-09-30", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "CB-1 and Autonomy-1", "domains": [ "bio_chem", "autonomy" ], "determination": "precautionary", "scope": "weights", "statement": "We do treat Sonnet 5.5 as meeting our CB-1 and Autonomy-1 thresholds and apply the corresponding mitigations from our RSP.", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "unpinnable": true, "verified_at": "2026-10-07", "research_confidence": "medium" } ] }, { "id": "model-apodex-apodex-1-1-mini", "source_id": null, "name": "Apodex 1.1 Mini", "developer": "Apodex AI", "url": "https://huggingface.co/apodex/Apodex-1.1-mini", "description": "Apodex AI's ungated 35B-total, 3B-active mixture-of-experts multimodal reasoning model, fine-tuned from Qwen3.5-35B-A3B for long-horizon research work with native function calling, published as Apache-2.0 open weights with official FP8, NVFP4, GPTQ-Int4, and GGUF exports alongside the FrontierAgent execution harness, and served as the Deep Research and Deep Solve tiers of the Apodex platform and through OpenRouter.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Apodex 1.1 Mini", "description": "Reasoning-first model for complex, long-horizon research tasks that works directly with files, data, code, and tools to produce verifiable deliverables", "family": "apodex", "release_date": "2026-08-24", "last_updated": "2026-08-26", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [ { "label": "Official release blog", "url": "https://www.apodex.com/blog/apodex-1.1-scaling-agentic-intelligence-for-complex-work" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/apodex/Apodex-1.1-mini" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships a LICENSE file carrying the verbatim Apache License 2.0 text, and the model card's own License section states that Apodex-1.1 is released under Apache 2.0. The grant covers the released weights and carries no revenue threshold, use restriction, or separate model-as-a-service licence, so the release is open source. The base checkpoint, Qwen/Qwen3.5-35B-A3B, is itself Apache-2.0, so the fine-tune inherits a compatible grant. Apodex's own hosted service is governed separately by its Commercial Terms of Service, which is a service contract and does not qualify the weight licence.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official apodex/Apodex-1.1-mini repository: Safetensors weights, tokenizer, chat template, preprocessor and video-preprocessor configuration, and the official FP8, NVFP4, GPTQ-Int4, and GGUF exports and their derivatives", "kind": "web_terms", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/LICENSE", "verified_at": "2026-10-01" }, { "license_id": "Apache-2.0", "scope": "Pinned model card at the reviewed repository commit, whose License section states the release terms and whose YAML front matter declares license: apache-2.0 and the Qwen/Qwen3.5-35B-A3B base model", "kind": "web_terms", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/README.md", "verified_at": "2026-10-01" } ], "access_boundary": "This record represents the apodex/Apodex-1.1-mini weight release published on Hugging Face at commit 4e4f109a5e32d1306dc6c75e4995a09c8b881811, the 35B-total, 3B-active multimodal mixture-of-experts checkpoint of the Apodex 1.1 generation. The unreleased full-size Apodex 1.1 model behind the Deep Discover tier, the Apodex 1.0 and 1.0-mini checkpoints and their 0.8B, 2B, and 4B SFT siblings, the official FP8, NVFP4, GPTQ-Int4, and GGUF format exports, the FrontierAgent and AgentCore harnesses, the Apodex Deep Research, Deep Solve, and Deep Discover product tiers, and OpenRouter's hosted route for the same weights are separate boundaries.", "strengths": [ "Ungated weights in an official Apodex Hugging Face organization, shipping the verbatim Apache-2.0 LICENSE file in the repository and stating the same terms in the card", "Official FP8, NVFP4, GPTQ-Int4, and GGUF exports published by the developer on the same organization, so a consumer-GPU path needs no third-party repackaging", "First-party model card publishes the recommended sampling parameters, the 262,144-token context and 32,768-token output budget, exact SGLang and vLLM launch commands with the Qwen3 tool-call and reasoning parsers, and a complete tool-using agent example", "The FrontierAgent harness and the Apodex 1.1 technical report are published under the same organization, so the execution environment the model is trained for is inspectable rather than described only in prose", "First-party documentation ties the 35B checkpoint to the Deep Research and Deep Solve platform tiers, and OpenRouter documents an independent hosted route" ], "tradeoffs": [ "Only the mini checkpoint is open; the full-size Apodex 1.1 model behind the Deep Discover tier is not published, so the generation's top capability is not obtainable as weights", "The base is a Qwen3.5 checkpoint, so the release inherits that lineage's identity rather than standing as an independently pretrained model", "The model card's benchmark results are reported for the Agent Team system configuration rather than for the bare checkpoint, so the published scores do not isolate the weights", "No deprecation or versioning posture is published for the line, and release identity rests on repository commits with no dated snapshot names" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.48 }, "evidence": [ { "kind": "web", "label": "Official Apodex 1.1 Mini model card: license, base model, sampling parameters, SGLang and vLLM launch commands, and the agentic usage example", "url": "https://huggingface.co/apodex/Apodex-1.1-mini", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Pinned model card at the reviewed repository commit, carrying the licence grant and the declared base model", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/README.md", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Apache-2.0 LICENSE file shipped in the weights repository", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/LICENSE", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Pinned model configuration giving the 262,144-token context, 40-layer linear/full attention layout, 8-of-256 expert routing, and the vision and video preprocessor token identifiers", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/config.json", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official Apodex 1.1 release announcement: the 35B mini as the locally deployable checkpoint, the FrontierAgent harness, and the full model as the complete capability envelope", "url": "https://www.apodex.com/blog/apodex-1.1-scaling-agentic-intelligence-for-complex-work", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official Apodex 1.1 technical report", "url": "https://www.apodex.com/pdf/20260824", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official research page naming the 35B Apodex Deep Research model and its Agent Team depth tiers", "url": "https://www.apodex.com/research", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official API pricing page listing the Deep Research 1.1, Deep Solve 1.1, and Deep Discover 1.1 Preview platform tiers", "url": "https://www.apodex.com/api", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official FrontierAgent repository: the execution harness the model card recommends for running the mini locally", "url": "https://github.com/ApodexAI/FrontierAgent", "verified_at": "2026-10-01" }, { "kind": "web", "label": "OpenRouter model page documenting a hosted route for the same release", "url": "https://openrouter.ai/apodex/apodex-1.1-mini", "verified_at": "2026-10-01" } ], "metadata_verified_at": "2026-10-01", "verified_at": "2026-10-01" }, { "id": "model-arcee-ai-trinity-large-preview", "source_id": "arcee-ai/trinity-large-preview", "name": "Trinity Large Preview", "developer": "Arcee AI", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview", "description": "Arcee AI's 398-billion-parameter sparse MoE preview with 13B active parameters, published as OpenMDW-1.1 open weights while undergoing further post-training.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Trinity Large Preview", "description": "Lightly post-trained 398B MoE chat model for creative work, long-context prompts, and tool-using agents", "family": "trinity", "release_date": "2026-01-27", "last_updated": "2026-05-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 524288, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": "OpenMDW-1.1", "links": [ { "label": "Model card", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview", "type": "model_card" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/trinity-large", "type": "announcement" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/blob/main/LICENSE", "type": "license" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview", "format": "safetensors" } ] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the OpenMDW License Agreement 1.1, which grants dealing in the model materials without restriction and imposes no obligations on outputs. Vendor blog and docs pages still carry older Apache-2.0 statements superseded by the May 2026 license migration; the file governs.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "Model weights and materials in the Trinity-Large-Preview repository", "kind": "web_terms", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/blob/d8e126719c8bf2b3ac5a3551bd1f07fcee767a90/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the January 2026 Trinity-Large-Preview release at Hugging Face commit d8e126719c8bf2b3ac5a3551bd1f07fcee767a90 (last modified 2026-05-28), a lightly post-trained checkpoint distinct from the Thinking release. The Base and TrueBase checkpoints, the Thinking release, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 398B MoE checkpoint with modeling code and five official quantization repos", "First-party preview API plus exact-model OpenRouter hosting" ], "tradeoffs": [ "Explicitly an evolving preview under active post-training, not a stable line", "Vendor blog and docs still carry superseded Apache-2.0 statements" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 4, "documentation_provenance": 8, "overall": 6.72 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Trinity Large launch announcement (January 2026)", "url": "https://www.arcee.ai/blog/trinity-large", "verified_at": "2026-09-18" }, { "kind": "web", "label": "OpenRouter model page for the exact preview model", "url": "https://openrouter.ai/arcee-ai/trinity-large-preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/arcee-ai/trinity-large-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-arcee-ai-trinity-large-thinking", "source_id": "arcee-ai/trinity-large-thinking", "name": "Trinity Large Thinking", "developer": "Arcee AI", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", "description": "Arcee AI's 398-billion-parameter reasoning-optimized release with extended chain-of-thought traces, published as OpenMDW-1.1 open weights with managed API hosting.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Trinity Large Thinking", "description": "Reasoning-optimized 398B MoE agent model with extended thinking for long-horizon and multi-turn tool use", "family": "trinity", "release_date": "2026-04-01", "last_updated": "2026-05-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 524288, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": "OpenMDW-1.1", "links": [ { "label": "Model card", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", "type": "model_card" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/trinity-large-thinking", "type": "announcement" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/main/LICENSE", "type": "license" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", "format": "safetensors" } ] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the OpenMDW License Agreement 1.1, which grants dealing in the model materials without restriction and imposes no obligations on outputs. Vendor blog and docs pages still carry older Apache-2.0 statements superseded by the May 2026 license migration; the file governs.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "Model weights and materials in the Trinity-Large-Thinking repository", "kind": "web_terms", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/dc6a99b6b74202880e31491a445a94884c97890c/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the April 2026 Trinity-Large-Thinking release at Hugging Face commit dc6a99b6b74202880e31491a445a94884c97890c (last modified 2026-05-28), the official reasoning successor of the Preview checkpoint. The Preview, Base, and TrueBase checkpoints, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 398B reasoning checkpoint with explicit thinking traces and four quantization repos", "First-party API plus exact-model OpenRouter hosting as the official Large release" ], "tradeoffs": [ "Vendor blog and docs still carry superseded Apache-2.0 statements", "398B dense-class parameters need datacenter-class multi-GPU serving" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 7.48 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Trinity Large Thinking launch announcement (April 2026)", "url": "https://www.arcee.ai/blog/trinity-large-thinking", "verified_at": "2026-09-18" }, { "kind": "web", "label": "OpenRouter model page for the exact thinking model", "url": "https://openrouter.ai/arcee-ai/trinity-large-thinking", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/arcee-ai/trinity-large-thinking.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-arcee-ai-trinity-mini", "source_id": "arcee-ai/trinity-mini", "name": "Trinity Mini", "developer": "Arcee AI", "url": "https://huggingface.co/arcee-ai/Trinity-Mini", "description": "Arcee AI's 26-billion-parameter MoE reasoning release with 3B active parameters, published as OpenMDW-1.1 open weights with managed API hosting.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Trinity Mini", "description": "Reasoning-tuned 26B MoE model with 3B active parameters for agents, tools, and multi-step workloads", "family": "trinity", "release_date": "2025-12-01", "last_updated": "2026-05-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": "OpenMDW-1.1", "links": [ { "label": "Model card", "url": "https://huggingface.co/arcee-ai/Trinity-Mini", "type": "model_card" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/the-trinity-manifesto", "type": "announcement" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/blob/main/LICENSE", "type": "license" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Mini", "format": "safetensors" } ] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the OpenMDW License Agreement 1.1, which grants dealing in the model materials without restriction and imposes no obligations on outputs. Vendor blog and docs pages still carry older Apache-2.0 statements superseded by the May 2026 license migration; the file governs.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "Model weights and materials in the Trinity-Mini repository", "kind": "web_terms", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/blob/cdb81e8130815fec299cccc8f8700004b8366e1e/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the December 2025 Trinity Mini release at Hugging Face commit cdb81e8130815fec299cccc8f8700004b8366e1e (last modified 2026-05-28), a fully post-trained reasoning line. Base and pre-anneal checkpoints, the Nano line, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 26B MoE checkpoint with GGUF, FP8, W4A16, and NVFP4 variants", "First-party API with free tier plus exact-model OpenRouter hosting and 128K context" ], "tradeoffs": [ "Vendor launch blog still carries the superseded Apache-2.0 statement", "Version identity rests on repository commits with planned distillations ahead" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.74 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/arcee-ai/Trinity-Mini", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Trinity manifesto announcement (December 2025)", "url": "https://www.arcee.ai/blog/the-trinity-manifesto", "verified_at": "2026-09-18" }, { "kind": "web", "label": "OpenRouter model page for the exact mini model", "url": "https://openrouter.ai/arcee-ai/trinity-mini", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/arcee-ai/trinity-mini.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-arcee-ai-trinity-nano-preview", "source_id": "arcee-ai/trinity-nano-preview", "name": "Trinity Nano Preview", "developer": "Arcee AI", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", "description": "Arcee AI's 6-billion-parameter experimental MoE preview with 1B active parameters, published as OpenMDW-1.1 download-only weights.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Trinity Nano Preview", "description": "Experimental chat-tuned 6B MoE model with 1B active parameters for low-resource chat and instruction following", "family": "trinity", "release_date": "2025-12-01", "last_updated": "2026-05-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": "OpenMDW-1.1", "links": [ { "label": "Model card", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", "type": "model_card" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/the-trinity-manifesto", "type": "announcement" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/blob/main/LICENSE", "type": "license" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", "format": "safetensors" } ] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the OpenMDW License Agreement 1.1, which grants dealing in the model materials without restriction and imposes no obligations on outputs. Vendor blog and docs pages still carry older Apache-2.0 statements superseded by the May 2026 license migration; the file governs.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "Model weights and materials in the Trinity-Nano-Preview repository", "kind": "web_terms", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/blob/b576b32b8bba7f5218312c9509e494537f8d282d/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the December 2025 Trinity Nano Preview experimental release at Hugging Face commit b576b32b8bba7f5218312c9509e494537f8d282d (last modified 2026-05-28), explicitly download-only with no hosted serving. The Mini line, a future graduated Nano line, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 6B edge-class checkpoint with GGUF, MLX, FP8, W4A16, and NVFP4 variants", "Runs on edge-class hardware with Transformers, vLLM, llama.cpp, and LM Studio paths" ], "tradeoffs": [ "Explicitly experimental with no hosted serving anywhere by developer statement", "Vendor pages still carry superseded Apache-2.0 statements; graduation may supersede this line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 4, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 6.88 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Trinity manifesto announcement (December 2025)", "url": "https://www.arcee.ai/blog/the-trinity-manifesto", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/arcee-ai/trinity-nano-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-baai-bge-m3", "source_id": "baai/bge-m3", "name": "BGE M3", "developer": "BAAI", "url": "https://huggingface.co/BAAI/bge-m3", "description": "BAAI's January 2024 multilingual embedding release combining dense, sparse, and multi-vector retrieval in one 568M-parameter checkpoint with 8K context and 100-plus-language coverage, published under the MIT licence with weights on BAAI's Hugging Face organization and documentation on the BGE site. No developer API or managed host serves this checkpoint. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "embedding_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "BGE M3", "description": "Multilingual embedding model combining dense, sparse, and multi-vector retrieval", "family": "bge", "release_date": "2024-01-30", "last_updated": "2024-01-30", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text", "embedding" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 8192, "input": null, "output": 0 }, "reported_open_weights": true, "reported_license": "MIT", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/BAAI/bge-m3" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The BAAI/bge-m3 model card declares license mit in its frontmatter and the repository metadata carries the same MIT tag; the repository ships no LICENSE file of its own, so the grant is the standard MIT text. No separate model-use terms were found on the pages read.", "license_evidence": [ { "license_id": "MIT", "scope": "BGE-M3 weights published in BAAI's Hugging Face repository under the MIT licence declared by the model card", "kind": "web_terms", "url": "https://opensource.org/licenses/MIT", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the BGE-M3 embedding checkpoint. The BGE rerankers, the bge-large/bge-base/bge-small line, the multilingual-Gemma BGE releases, and the FlagEmbedding toolkit repository are separate boundaries.", "strengths": [ "One checkpoint serves dense, sparse, and multi-vector retrieval across more than one hundred languages with 8,192-token context", "Permissive MIT grant declared both in the model card and the repository metadata, with weights in PyTorch and ONNX formats" ], "tradeoffs": [ "No developer API or managed host serves this checkpoint, so running it is entirely self-operated", "The repository carries no versioned release beyond the single checkpoint, so line updates arrive as new repositories rather than revisions" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.92 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/59aa52dbc09727fa6ead433b0285970ebc05ba82/models/baai/bge-m3.toml", "verified_at": "2026-10-13" }, { "kind": "web", "label": "BAAI BGE-M3 weight repository: MIT licence, multilingual embedding", "url": "https://huggingface.co/BAAI/bge-m3", "verified_at": "2026-10-13" }, { "kind": "web", "label": "BGE documentation: BGE is developed by the Beijing Academy of Artificial Intelligence", "url": "https://bge-model.com/", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/BAAI/bge-m3", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-baidu-ernie-4-5-21b-a3b", "source_id": "baidu/ernie-4.5-21b-a3b", "name": "ERNIE 4.5 21B A3B", "developer": "Baidu", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT", "description": "Baidu's June 30, 2025 open release of ERNIE 4.5 as a 21B-total, 3B-active mixture-of-experts text checkpoint with 131K context and tool calling, published under Apache 2.0 with safetensors weights on Baidu's Hugging Face organization alongside a technical report. No developer API or managed host is documented for this exact checkpoint. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "ERNIE 4.5 21B A3B", "description": "Open ERNIE text MoE with 21B total and 3B active parameters", "family": "ernie", "release_date": "2025-06-30", "last_updated": "2025-06-30", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8000 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "Baidu's ERNIE 4.5 announcement states that all ten family models are publicly accessible under Apache 2.0, and the baidu/ERNIE-4.5-21B-A3B-PT repository carries an Apache License, Version 2.0 LICENSE file and declares license apache-2.0. The grant covers the published weights; no separate model-use terms were found on the pages read.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "ERNIE-4.5-21B-A3B-PT weights published in Baidu's Hugging Face repository", "kind": "web_terms", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT/blob/87db95487941cb39592ee0abca3b9155a6d19c5c/LICENSE", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the ERNIE-4.5-21B-A3B-PT post-trained checkpoint of the ERNIE 4.5 family. The Base-PT and Thinking variants, the 0.3B, 30B, 300B, and 424B siblings, the VL multimodal variants, the Qianfan ernie-4.5-turbo API models, and the ERNIE Bot product are separate boundaries.", "strengths": [ "Apache 2.0 grant stated both in Baidu's release announcement and in a LICENSE file inside the weight repository, with the repository metadata declaring the same licence", "Versioned post-trained checkpoint with safetensors weights, a model card, and a dated technical report, loadable through standard transformers and vLLM tooling" ], "tradeoffs": [ "No developer API or managed host is documented for this exact checkpoint, so serving it is entirely self-operated", "The 21B mixture-of-experts checkpoint still needs accelerator-class hardware despite only 3B active parameters per token" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 2, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.44 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/59aa52dbc09727fa6ead433b0285970ebc05ba82/models/baidu/ernie-4.5-21b-a3b.toml", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Baidu ERNIE 4.5 open-source announcement: June 30, 2025, ten variants, Apache 2.0", "url": "https://ernie.baidu.com/blog/posts/ernie4.5/", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Baidu ERNIE-4.5-21B-A3B-PT weight repository: apache-2.0 licence, safetensors, text generation", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-bfl-flux-2-dev", "source_id": null, "name": "FLUX.2 [dev]", "developer": "Black Forest Labs", "url": "https://huggingface.co/black-forest-labs/FLUX.2-dev", "description": "Black Forest Labs' 32B rectified flow transformer for text-to-image generation and single- or multi-reference image editing, published as gated open weights under the FLUX [dev] Non-Commercial License 2.0, with no first-party developer API for this checkpoint.", "model_type": "image_generation_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "FLUX.2 [dev]", "description": "Black Forest Labs' open-weight image generation and editing model, the 32 billion parameter member of the FLUX.2 family", "family": "flux", "release_date": "2025-11-25", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": null, "reasoning": null, "structured_output": null, "temperature": null, "tool_call": null }, "limits": { "context": null, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "flux-non-commercial-license", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/black-forest-labs/FLUX.2-dev" } ] }, "licenses": [ "LicenseRef-FLUX-Dev-Non-Commercial-2.0" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The weights, parameters, and inference code for every FLUX.x [dev] model are governed by the FLUX [dev] Non-Commercial License 2.0, published on the company's own legal pages with a version number and a last-updated date of November 25, 2025, and named for FLUX.2 [dev] in its own definition of the models it covers. It grants use, distribution, and derivative works only for a Non-Commercial Purpose, which its own text excludes from revenue-generating activity, end-user-facing deployment, and training a competing model; it requires content filters or a pre-distribution review of every output under section 2(e) together with AI disclosure; and the company may terminate it on notice under section 8(b). Outputs may be used commercially, so the restriction falls on operating the model rather than on the images it produces. The repository ships no LICENSE file, so the published web terms are the governing text and the Hugging Face licence tag is a card label rather than evidence. The licence names Black Forest Labs Inc. while the imprint names BFL GmbH, a difference the organization's own pages do not reconcile. Commercial self-hosting is sold separately under Self-Hosted Commercial License Terms, which govern a negotiated arrangement rather than this published release.", "license_evidence": [ { "license_id": "LicenseRef-FLUX-Dev-Non-Commercial-2.0", "scope": "FLUX.2 [dev] weights, parameters, and inference code, including the content-filter and output-disclosure conditions in sections 2 and 8", "kind": "web_terms", "url": "https://bfl.ai/legal/non-commercial-license-terms", "verified_at": "2026-10-01" } ], "access_boundary": "This record represents the FLUX.2 [dev] open-weight release announced November 25, 2025 and served from the gated black-forest-labs/FLUX.2-dev repository. The FLUX.2 [klein], [flex], [pro], and [max] API tiers, the FLUX 3 image and video models, the FLUX 3 Action world action model, the FLUX 2 [klein] training path, the FLUX Tools endpoints, and third-party derivatives and quantizations are separate boundaries.", "strengths": [ "The governing licence is published on an official page with a version number and a revision date, and names the models it covers", "Documented deployment paths through Diffusers, ComfyUI, the single-file format, and a 4-bit quantization for consumer GPUs" ], "tradeoffs": [ "The weights are gated behind an accepted licence agreement, and the licence restricts operating the model to non-commercial purposes while requiring content filters on every deployment", "No first-party developer API serves this checkpoint, so commercial reach depends on a separately negotiated licence or a third-party host" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 6, "deployment_portability": 8, "serving_reach": 3, "lifecycle_transparency": 6, "documentation_provenance": 9, "overall": 6.18 }, "evidence": [ { "kind": "web", "label": "Developer-published FLUX.2 [dev] weight repository and model card", "url": "https://huggingface.co/black-forest-labs/FLUX.2-dev", "verified_at": "2026-10-01" }, { "kind": "web", "label": "FLUX [dev] Non-Commercial License 2.0: the models it covers, the non-commercial purpose, the content-filter condition, and revocability", "url": "https://bfl.ai/legal/non-commercial-license-terms", "verified_at": "2026-10-01" }, { "kind": "web", "label": "FLUX.2 announcement of November 25, 2025: the open-weight release of the FLUX.2 family", "url": "https://bfl.ai/blog/flux-2", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Developer FLUX.2 overview: the API tiers are [max], [pro], [flex], and [klein], so no first-party endpoint serves this checkpoint", "url": "https://docs.bfl.ai/flux_2/flux2_overview", "verified_at": "2026-10-01" }, { "kind": "web", "label": "fal's own model page hosting this release as fal-ai/flux-2-dev", "url": "https://fal.ai/models/fal-ai/flux-2-dev", "verified_at": "2026-10-01" } ], "metadata_verified_at": "2026-10-01", "verified_at": "2026-10-01" }, { "id": "model-bytedance-seed-seed-1-6", "source_id": "bytedance-seed/seed-1-6", "name": "Seed 1.6", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/blog/introduction-to-techniques-used-in-seed1-6", "description": "ByteDance's 230-billion-parameter Mixture-of-Experts release with 23B active parameters and 256K context, served as the doubao-seed-1-6 API snapshots with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 1.6", "description": "ByteDance Seed model for long-context reasoning, instruction following, and tool-assisted tasks", "family": "seed", "release_date": "2025-10-15", "last_updated": "2025-10-15", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 64000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of doubao-seed-1-6 snapshots with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the June 2025 Seed 1.6 line served as dated doubao-seed-1-6 API snapshots (250615, 251015) on Volcano Engine and BytePlus ModelArk. The Flash, Vision, and 1.8 lines, thinking variants, and both platforms themselves are separate boundaries.", "strengths": [ "Dated API snapshots with a signed Seed technical report and 256K context", "Two first-party serving platforms: Volcano Engine and BytePlus ModelArk" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 2, "deployment_portability": 1, "serving_reach": 8, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.18 }, "evidence": [ { "kind": "web", "label": "Official Seed 1.6 techniques blog (June 2025)", "url": "https://seed.bytedance.com/blog/introduction-to-techniques-used-in-seed1-6", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Volcano Engine model list with the dated seed-1-6 snapshots", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-1-6.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-1-6-flash", "source_id": "bytedance-seed/seed-1-6-flash", "name": "Seed 1.6 Flash", "developer": "ByteDance Seed", "url": "https://www.volcengine.com/docs/82379/1330310", "description": "ByteDance's speed-optimized Seed 1.6 line served as dated doubao-seed-1-6-flash API snapshots with 256K context and no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 1.6 Flash", "description": "Low-latency ByteDance Seed model for high-throughput chat, extraction, and lightweight tool use", "family": "seed", "release_date": "2025-08-28", "last_updated": "2025-08-28", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 32000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of doubao-seed-1-6-flash snapshots with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the Seed 1.6 Flash line served as dated doubao-seed-1-6-flash API snapshots (250615, 250828) on Volcano Engine. The base 1.6, Vision, and 1.8 lines and the platform itself are separate boundaries.", "strengths": [ "Two dated API snapshots document the line as a distinct speed tier rather than an alias", "256K context with tool calling and structured output on first-party endpoints" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 2, "deployment_portability": 1, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 3.9 }, "evidence": [ { "kind": "web", "label": "Official Volcano Engine model list with the dated flash snapshots", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-1-6-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-1-6-vision", "source_id": "bytedance-seed/seed-1-6-vision", "name": "Seed 1.6 Vision", "developer": "ByteDance Seed", "url": "https://www.volcengine.com/docs/82379/1330310", "description": "ByteDance's vision-specialized Seed 1.6 line for multimodal understanding and GUI tasks, served as dated API snapshots with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 1.6 Vision", "description": "ByteDance Seed multimodal model for image understanding, visual reasoning, and tool-assisted tasks", "family": "seed", "release_date": "2025-08-15", "last_updated": "2025-08-15", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 32000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of doubao-seed-1-6-vision snapshots with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the Seed 1.6 Vision line served as the dated doubao-seed-1-6-vision-250815 API snapshot on Volcano Engine. The base 1.6, Flash, and 1.8 lines and the platform itself are separate boundaries.", "strengths": [ "Dated API snapshot documents the line as a distinct vision specialization with GUI task handling", "Multimodal understanding across text, image, and video inputs" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 2, "deployment_portability": 1, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 3.9 }, "evidence": [ { "kind": "web", "label": "Official Volcano Engine model list with the dated vision snapshot", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-1-6-vision.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-1-8", "source_id": "bytedance-seed/seed-1-8", "name": "Seed 1.8", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/blog/official-release-of-seed1-8-a-generalized-agentic-model", "description": "ByteDance's December 2025 generalized agentic release with text and image input, served as the doubao-seed-1-8-251228 API snapshot with cookbook code and no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 1.8", "description": "ByteDance Seed model for multimodal reasoning, long-context analysis, and agent workflows", "family": "seed", "release_date": "2025-12-28", "last_updated": "2025-12-28", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 64000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model-weight license exists: no weights were released, and the cookbook repository's Apache-2.0 file covers code samples only. Use of the model is governed by platform API terms, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of doubao-seed-1-8-251228 with no published model-weight license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the December 2025 Seed 1.8 line served as the doubao-seed-1-8-251228 API snapshot on Volcano Engine, with the Apache-2.0 cookbook repository as its only published artifact. Earlier Seed lines and the platform itself are separate boundaries.", "strengths": [ "Dated launch announcement pins the December 2025 release with text and image input", "Apache-2.0 cookbook repository documents the release, pinned to its init commit" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No model-weight license of any kind; the Apache file covers code samples only" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 4, "deployment_portability": 3, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 5.24 }, "evidence": [ { "kind": "web", "label": "Official Seed 1.8 launch announcement (December 2025)", "url": "https://seed.bytedance.com/blog/official-release-of-seed1-8-a-generalized-agentic-model", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Volcano Engine model list with the dated 1.8 snapshot", "url": "https://www.volcengine.com/docs/82379/1330310", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-1-8.toml", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Apache-2.0 cookbook repository (code samples only, pinned init commit)", "url": "https://github.com/ByteDance-Seed/Seed-1.8", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-2-0-code", "source_id": "bytedance-seed/seed-2.0-code", "name": "Seed 2.0 Code", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "description": "ByteDance's dedicated Seed 2.0 coding release served through the Doubao app, TRAE, and the Volcano Engine API with no public weights.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 2.0 Code", "description": "ByteDance Seed coding model for multimodal software engineering and long-running agents", "family": "seed", "release_date": "2026-02-14", "last_updated": "2026-02-14", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service, which reserve all service intellectual property to the operator.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the Seed 2.0 Code line with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/6256/64903", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the February 2026 Seed 2.0 Code line served as Doubao-Seed-2.0-Code through TRAE and the Volcano Engine API. The Pro, Lite, and Mini siblings, later Seed generations, and all three platforms themselves are separate boundaries.", "strengths": [ "Dated February 2026 launch pins the line with an exact TRAE model entry", "Doubao app, TRAE, and full-series Volcano Engine API serving" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.94 }, "evidence": [ { "kind": "web", "label": "Official Seed 2.0 launch announcement (February 2026)", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official API cookbook repository (code samples only)", "url": "https://github.com/ByteDance-Seed/Seed2.0", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-2.0-code.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-2-0-lite", "source_id": "bytedance-seed/seed-2.0-lite", "name": "Seed 2.0 Lite", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "description": "ByteDance's omni-modal Seed 2.0 Lite release with unified text, image, video, and audio understanding, served through the Volcano Engine API with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 2.0 Lite", "description": "Cost-efficient ByteDance Seed 2.0 model for production chat, analysis, and structured generation", "family": "seed", "release_date": "2026-02-14", "last_updated": "2026-02-14", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 32000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service, which reserve all service intellectual property to the operator.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the Seed 2.0 Lite line with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/6256/64903", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the February 2026 Seed 2.0 Lite line, upgraded end of April 2026 with audio input, served through the Volcano Engine API. The Pro, Mini, and Code siblings, later Seed generations, and the platform itself are separate boundaries.", "strengths": [ "Dated snapshots (0215, 0428) document the line's April omni-modal upgrade", "Unified text, image, video, and audio understanding on first-party endpoints" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.06 }, "evidence": [ { "kind": "web", "label": "Official Seed 2.0 launch announcement (February 2026)", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-2.0-lite.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-2-0-mini", "source_id": "bytedance-seed/seed-2.0-mini", "name": "Seed 2.0 Mini", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "description": "ByteDance's throughput-optimized Seed 2.0 Mini release for high-concurrency workloads, served through the Volcano Engine API with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 2.0 Mini", "description": "Lightweight ByteDance Seed 2.0 model for low-latency multimodal reasoning and high-volume tasks", "family": "seed", "release_date": "2026-02-14", "last_updated": "2026-02-14", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 32000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service, which reserve all service intellectual property to the operator.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the Seed 2.0 Mini line with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/6256/64903", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the February 2026 Seed 2.0 Mini line served through the Volcano Engine API. The Pro, Lite, and Code siblings, later Seed generations, and the platform itself are separate boundaries.", "strengths": [ "Dated February 2026 launch pins the line with a dated 0215 deployment", "Throughput-optimized serving for high-concurrency workloads" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.8 }, "evidence": [ { "kind": "web", "label": "Official Seed 2.0 launch announcement (February 2026)", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-2.0-mini.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-2-0-pro", "source_id": "bytedance-seed/seed-2.0-pro", "name": "Seed 2.0 Pro", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "description": "ByteDance's Seed 2.0 flagship line for agentic work, served through Volcano Engine, BytePlus, Doubao, and TRAE with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 2.0 Pro", "description": "Flagship ByteDance Seed 2.0 model for complex multimodal reasoning and long-horizon agent workflows", "family": "seed", "release_date": "2026-02-14", "last_updated": "2026-02-14", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service, which reserve all service intellectual property to the operator.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the Seed 2.0 Pro line with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/6256/64903", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the February 2026 Seed 2.0 Pro flagship line served as doubao-seed-2.0-pro with dated 0215 and 260328 deployments. The Lite, Mini, and Code siblings, the 2.1 successor generation, and all four platforms themselves are separate boundaries.", "strengths": [ "Dated flagship launch with dated snapshot deployments across two platform generations", "Four first-party surfaces: Volcano Engine, BytePlus, Doubao, and TRAE" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 4.18 }, "evidence": [ { "kind": "web", "label": "Official Seed 2.0 launch announcement (February 2026)", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-2.0-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-2-1-pro", "source_id": "bytedance-seed/seed-2.1-pro", "name": "Seed 2.1 Pro", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "description": "ByteDance's June 2026 Seed 2.1 flagship line for real-world productivity, served as the doubao-seed-2-1-pro-260628 snapshot with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 2.1 Pro", "description": "Flagship ByteDance Seed 2.1 model for complex multimodal reasoning, coding, and agents", "family": "seed", "release_date": "2026-06-23", "last_updated": "2026-06-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 256000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service, which reserve all service intellectual property to the operator.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the Seed 2.1 Pro line with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/6256/64903", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the June 2026 Seed 2.1 Pro flagship line served as doubao-seed-2-1-pro-260628 with 256K context. The Turbo sibling tier, the 2.0 predecessor generation, and all platforms themselves are separate boundaries.", "strengths": [ "Dated June 2026 launch pins the line with a single dated snapshot and proactive eval disclosure", "First-party Volcano and Doubao serving with 256K context" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.44 }, "evidence": [ { "kind": "web", "label": "Official Seed 2.1 launch announcement (June 2026)", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-2.1-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-seed-2-1-turbo", "source_id": "bytedance-seed/seed-2.1-turbo", "name": "Seed 2.1 Turbo", "developer": "ByteDance Seed", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "description": "ByteDance's smaller Seed 2.1 tier balancing capability and cost, served as the doubao-seed-2-1-turbo-260628 snapshot with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Seed 2.1 Turbo", "description": "Faster ByteDance Seed 2.1 model for multimodal reasoning and latency-sensitive agent workflows", "family": "seed", "release_date": "2026-06-23", "last_updated": "2026-06-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 256000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model weights or license document exist for this API-only line; use is governed by the Volcano Engine and BytePlus platform terms of service, which reserve all service intellectual property to the operator.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the Seed 2.1 Turbo tier with no published model license", "kind": "web_terms", "url": "https://www.volcengine.com/docs/6256/64903", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the June 2026 Seed 2.1 Turbo tier served as doubao-seed-2-1-turbo-260628, a distinct smaller tier rather than an alias of Pro. The Pro sibling, the 2.0 predecessor generation, and all platforms themselves are separate boundaries.", "strengths": [ "Dated June 2026 launch pins the tier with its own snapshot and eval column", "First-party Volcano and Doubao serving for the exact turbo snapshot ID" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this tier", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 7, "overall": 4.3 }, "evidence": [ { "kind": "web", "label": "Official Seed 2.1 launch announcement (June 2026)", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/bytedance-seed/seed-2.1-turbo.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-bytedance-seed-ui-tars-1-5-7b", "source_id": null, "name": "UI-TARS-1.5-7B", "developer": "ByteDance Seed", "url": "https://huggingface.co/ByteDance-Seed/UI-TARS-1.5-7B", "description": "ByteDance Seed's UI-TARS-1.5-7B, the open 7B checkpoint of the UI-TARS-1.5 GUI agent line: a vision-language model that reads screenshots and writes a reasoning thought followed by its next interface action as text, published as open weights under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "UI-TARS-1.5-7B", "description": "Open-source multimodal agent built on a vision-language model for GUI, computer-use, and game tasks, released as a 7B checkpoint", "family": "ui-tars", "release_date": "2025-04-17", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": null, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [ { "label": "Official release announcement", "url": "https://seed-tars.com/1.5/" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/ByteDance-Seed/UI-TARS-1.5-7B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card metadata declares Apache-2.0. The repository ships no separate LICENSE file (the raw LICENSE path returns 404 on 2026-10-06) and its file list holds only weights, tokenizer, and configuration files, so the grant rests on the card metadata alone. The card names no base model.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "UI-TARS-1.5-7B repository weights, Safetensors checkpoints, tokenizer, and preprocessor configuration", "kind": "web_terms", "url": "https://huggingface.co/ByteDance-Seed/UI-TARS-1.5-7B", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the UI-TARS-1.5-7B weight release in the ByteDance-Seed/UI-TARS-1.5-7B repository, announced 17 April 2025. The model card states a stronger UI-TARS-1.5 model is offered only as early research access on request, and that model, the UI-TARS-desktop application, the paper's earlier UI-TARS checkpoints, and gateway-hosted copies are separate boundaries. Its output is text that names GUI actions; the record covers the model, not any harness that executes those actions.", "strengths": [ "Ungated Safetensors weights under Apache-2.0 declared on the model card, with code and desktop application linked from the card", "Dated first-party announcement with a stated scope: the 7B checkpoint targets general computer use rather than games" ], "tradeoffs": [ "The Apache-2.0 grant rests on model-card metadata with no LICENSE file in the repository", "The strongest UI-TARS-1.5 model is not released as open weights, and no first-party hosted API for this checkpoint was found in the sources read" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 3, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.4 }, "evidence": [ { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/ByteDance-Seed/UI-TARS-1.5-7B", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official UI-TARS-1.5 announcement", "url": "https://seed-tars.com/1.5/", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-cohere-command-a-03-2025", "source_id": "cohere/command-a-03-2025", "name": "Command A", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025", "description": "A gated downloadable text language-model release from Cohere Labs whose weights are limited to noncommercial use under CC-BY-NC-4.0 with an incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Command A", "description": "Cohere command model for multilingual enterprise agents, tools, and chat", "family": "command-a", "release_date": "2025-03-13", "last_updated": "2025-03-13", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 8000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases the c4ai-command-a-03-2025 weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which Cohere's license page incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; Cohere's hosted command-a-03-2025 is offered under separate commercial terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "Cohere Labs Command A (c4ai-command-a-03-2025) weights, tokenizer, and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/c4ai-cc-by-nc-license", "verified_at": "2026-09-15" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the CohereLabs/c4ai-command-a-03-2025 weight release. Cohere's hosted command-a-03-2025 API model and its cloud-marketplace copies, downstream quantizations and fine-tunes, and later Command A releases such as command-a-reasoning-08-2025 remain separate model or service boundaries.", "strengths": [ "Cohere publishes Safetensors shards with an index, tokenizer files, and a signature file with verification instructions", "The official model card documents Transformers, vLLM, SGLang, and Docker Model Runner paths, and Cohere documents a first-party API plus Azure AI Foundry and Oracle OCI hosting" ], "tradeoffs": [ "Artifact access is gated and the noncommercial license prevents production reuse of the weights without separate permission", "The model card sets a 128K default context in its configuration and hosted availability varies by platform" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.84 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Cohere model list with API identifier and platform availability", "url": "https://docs.cohere.com/docs/models", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/cohere/command-a-03-2025.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-cohere-command-a-plus-05-2026", "source_id": "cohere/command-a-plus-05-2026", "name": "Command A Plus", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/command-a-plus-05-2026-w4a4", "description": "Cohere's 218-billion-parameter Mixture-of-Experts Command A+ release with 25B active parameters, published as ungated Apache-2.0 weights in three precisions with managed API and Azure hosting.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Command A Plus", "description": "Cohere's stronger command model for multilingual agents and enterprise workflows", "family": "command-a", "release_date": "2026-05-20", "last_updated": "2026-06-09", "knowledge_cutoff": "2025-04-01", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 64000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card, launch announcement, and docs page declare Command A+ available under the Apache 2.0 license with no acceptable-use addendum or noncommercial cap, unlike the CC-BY-NC Command A line. No LICENSE file ships in the weight repositories, so the grant rests on those concordant first-party declarations pointing at the standard Apache-2.0 text.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Command A+ weights as declared on the official model card and docs page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the May 2026 Command A+ release across its three published precisions (bf16, fp8, w4a4), together with the hosted command-a-plus-05-2026 API model and the Azure w4a4 hosting. The Command A base line, other Command variants, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated weights in three published precisions under a declared Apache-2.0 grant with per-quant serving minima", "First-party API plus a documented Azure AI Foundry hosting entry for the exact w4a4 artifact" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card, blog, and docs declarations", "218B total parameters need accelerator-class serving outside the W4A4 path" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.34 }, "evidence": [ { "kind": "web", "label": "Official model card declaring the Apache-2.0 grant", "url": "https://huggingface.co/CohereLabs/command-a-plus-05-2026-w4a4", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Command A+ docs page declaring the Apache-2.0 grant", "url": "https://docs.cohere.com/docs/command-a-plus", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Cohere model documentation with the Azure hosting entry", "url": "https://docs.cohere.com/docs/models", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-a-plus-05-2026.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-command-a-reasoning-08-2025", "source_id": "cohere/command-a-reasoning-08-2025", "name": "Command A Reasoning", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/command-a-reasoning-08-2025", "description": "Cohere's 111-billion-parameter Command A Reasoning release with a reasoning toggle, published as gated weights under CC-BY-NC-4.0 with a hosted Chat API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Command A Reasoning", "description": "Cohere reasoning model for multilingual enterprise agents, tools, and complex workflows", "family": "command-a", "release_date": "2025-08-21", "last_updated": "2025-08-21", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 32000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-reasoning-08-2025" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs command-a-reasoning-08-2025 weights and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/command-a-reasoning-08-2025 weight release of August 2025, a distinct reasoning line rather than a mode of the base Command A release, together with the hosted command-a-reasoning-08-2025 Chat API model. Other Command A variants, Cohere's hosted API platform, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 111B reasoning weight release under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Cohere's docs serve command-a-reasoning-08-2025 on the Chat API with a 256K context window" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No third-party host documents this exact model; serving beyond Cohere is self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.56 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/command-a-reasoning-08-2025", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Command A Reasoning documentation", "url": "https://docs.cohere.com/docs/command-a-reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-a-reasoning-08-2025.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-command-a-translate-08-2025", "source_id": "cohere/command-a-translate-08-2025", "name": "Command A Translate", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/command-a-translate-08-2025", "description": "Cohere's 111-billion-parameter translation-specialized Command A release across 23 languages, published as gated weights under CC-BY-NC-4.0 with a hosted Chat API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Command A Translate", "description": "Translation model for multilingual conversion, localization, and cross-language workflows", "family": "command-a", "release_date": "2025-08-28", "last_updated": "2025-08-28", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 8000, "input": null, "output": 8000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-translate-08-2025" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs command-a-translate-08-2025 weights and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/command-a-translate-08-2025 weight release of August 2025, a distinct translation-specialized line, together with the hosted command-a-translate-08-2025 Chat API model. Other Command A variants, Cohere's hosted API platform, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 111B translation weight release under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Cohere's docs serve command-a-translate-08-2025 on the Chat API across 23 languages" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No third-party host documents this exact model; serving beyond Cohere is self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.56 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/command-a-translate-08-2025", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Command A Translate documentation", "url": "https://docs.cohere.com/docs/command-a-translate", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-a-translate-08-2025.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-command-a-vision-07-2025", "source_id": "cohere/command-a-vision-07-2025", "name": "Command A Vision", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/command-a-vision-07-2025", "description": "Cohere's 112-billion-parameter multimodal Command A release accepting image and text input, published as gated weights under CC-BY-NC-4.0 with a hosted Chat API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Command A Vision", "description": "Cohere vision model for multilingual document analysis, OCR, and image understanding", "family": "command-a", "release_date": "2025-07-31", "last_updated": "2025-07-31", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 8000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-vision-07-2025" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs command-a-vision-07-2025 weights and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/command-a-vision-07-2025 weight release of July 2025 with image and text input and text output, together with the hosted command-a-vision-07-2025 Chat API model. Other Command A variants, Cohere's hosted API platform, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 112B vision-language weights under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Cohere's docs serve command-a-vision-07-2025 on the Chat API with image input" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No third-party host documents this exact model; serving beyond Cohere is self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 6.42 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/command-a-vision-07-2025", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Command A Vision documentation", "url": "https://docs.cohere.com/docs/command-a-vision", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-a-vision-07-2025.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-command-r-08-2024", "source_id": "cohere/command-r-08-2024", "name": "Command R", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024", "description": "Cohere's 32-billion-parameter Command R refresh for retrieval-augmented generation and tool use, published as gated weights under CC-BY-NC-4.0 with a hosted Cohere API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Command R", "description": "Cohere retrieval model for long-context chat and enterprise RAG workflows", "family": "command-r", "release_date": "2024-08-30", "last_updated": "2024-08-30", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs c4ai-command-r-08-2024 weights, tokenizer, and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/c4ai-command-r-08-2024 weight release of August 2024, together with the hosted command-r-08-2024 API model. Earlier Command R snapshots, the R+ and R7B lines, Cohere's hosted API platform and cloud-marketplace copies, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated weight release under explicit CC-BY-NC-4.0 plus acceptable-use terms, with tokenizer and chat templates", "Cohere's own docs list command-r-08-2024 as a Live API model with a 128K context window" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No third-party host documents this exact snapshot; serving beyond Cohere is self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Cohere model documentation listing command-r-08-2024", "url": "https://docs.cohere.com/docs/models", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-r-08-2024.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-command-r-plus-08-2024", "source_id": "cohere/command-r-plus-08-2024", "name": "Command R+", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024", "description": "Cohere's 104-billion-parameter Command R+ refresh for demanding retrieval and tool workflows, published as gated weights under CC-BY-NC-4.0 with a hosted Cohere API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Command R+", "description": "Cohere's RAG workhorse for long-context enterprise search and tool use", "family": "command-r", "release_date": "2024-08-30", "last_updated": "2024-08-30", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs c4ai-command-r-plus-08-2024 weights, tokenizer, and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/c4ai-command-r-plus-08-2024 weight release of August 2024, together with the hosted command-r-plus-08-2024 API model. Earlier Command R+ snapshots, the R and R7B lines, Cohere's hosted API platform and cloud-marketplace copies, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 104B weight release under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Cohere's own docs list command-r-plus-08-2024 as a Live API model with a 128K context window" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No third-party host documents this exact snapshot; serving beyond Cohere is self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Cohere model documentation listing command-r-plus-08-2024", "url": "https://docs.cohere.com/docs/models", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-r-plus-08-2024.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-command-r7b-12-2024", "source_id": "cohere/command-r7b-12-2024", "name": "Command R7B", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-12-2024", "description": "Cohere's 7-billion-parameter small fast Command refresh, published as gated weights under CC-BY-NC-4.0 with a hosted Cohere API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Command R7B", "description": "Cohere retrieval model for long-context chat and enterprise RAG workflows", "family": "command-r", "release_date": "2024-12-02", "last_updated": "2024-12-02", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-12-2024" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs c4ai-command-r7b-12-2024 weights, tokenizer, and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/c4ai-command-r7b-12-2024 weight release of December 2024, together with the hosted command-r7b-12-2024 API model. The R and R+ lines, the Arabic sibling, Cohere's hosted API platform, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 7B weight release under explicit CC-BY-NC-4.0 plus acceptable-use terms, sized for smaller deployments", "Cohere's own docs list command-r7b-12-2024 as a Live API model with a 128K context window" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No third-party host documents this exact snapshot; serving beyond Cohere is self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-12-2024", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Cohere model documentation listing command-r7b-12-2024", "url": "https://docs.cohere.com/docs/models", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Command R7B launch announcement (December 2024)", "url": "https://cohere.com/blog/command-r7b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-r7b-12-2024.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-command-r7b-arabic-02-2025", "source_id": "cohere/command-r7b-arabic-02-2025", "name": "Command R7B Arabic", "developer": "Cohere", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025", "description": "Cohere's Arabic-optimized Command R7B release, published as gated weights under CC-BY-NC-4.0 with playground serving.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Command R7B Arabic", "description": "Open Command R model optimized for Arabic enterprise chat, RAG, and cultural knowledge", "family": "command-r", "release_date": "2025-02-27", "last_updated": "2025-02-27", "knowledge_cutoff": "2024-06-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs c4ai-command-r7b-arabic-02-2025 weights, tokenizer, and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/c4ai-command-r7b-arabic-02-2025 weight release of February 2025 for Arabic and English workloads, with playground serving under the exact model ID. The base R7B line, other Command lines, and downstream quantizations and fine-tunes remain separate boundaries. Cohere's changelog announces this release under the base R7B ID by mistake; the repository, card, paper, and blog fix the true identity recorded here.", "strengths": [ "Gated Arabic-optimized weight release under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Cohere's February 2025 launch announcement pins the release with its research paper" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No models-table row or third-party host documents this exact ID; managed serving is playground-only" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.44 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Command R7B Arabic launch announcement (February 2025)", "url": "https://cohere.com/blog/command-r7b-arabic", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/command-r7b-arabic-02-2025.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-c4ai-aya-expanse-32b", "source_id": "cohere/c4ai-aya-expanse-32b", "name": "Aya Expanse 32B", "developer": "Cohere Labs", "url": "https://huggingface.co/CohereLabs/aya-expanse-32b", "description": "Cohere Labs' 32-billion-parameter multilingual research release, published as gated weights under CC-BY-NC-4.0 with a hosted Chat API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Aya Expanse 32B", "description": "Open multilingual model optimized for generation across 23 languages", "family": null, "release_date": "2024-10-24", "last_updated": "2024-10-24", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": "CC-BY-NC-4.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-expanse-32b" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs aya-expanse-32b weights and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/aya-expanse-32b weight release of October 2024, together with the hosted c4ai-aya-expanse-32b Chat API model. The 8B sibling, the Aya Vision lines, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 32B multilingual weight release under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Cohere's Aya docs serve c4ai-aya-expanse-32b on the Chat API with a 128K context window" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No changelog tracks the line beyond the dated paper and repository" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.76 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/aya-expanse-32b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Cohere Aya Expanse documentation serving c4ai-aya-expanse-32b", "url": "https://docs.cohere.com/docs/aya-expanse", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/c4ai-aya-expanse-32b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-c4ai-aya-expanse-8b", "source_id": "cohere/c4ai-aya-expanse-8b", "name": "Aya Expanse 8B", "developer": "Cohere Labs", "url": "https://huggingface.co/CohereLabs/aya-expanse-8b", "description": "A multilingual research model distributed as downloadable Safetensors weights under noncommercial terms plus Cohere Labs' acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Aya Expanse 8B", "description": "Compact open multilingual model optimized for generation across 23 languages", "family": null, "release_date": "2024-10-24", "last_updated": "2024-10-24", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 8000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": "CC-BY-NC-4.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-expanse-8b" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card limits the release to noncommercial use and requires compliance with the Cohere Labs Acceptable Use Policy. Public weights therefore do not make this an open-source model distribution.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "Aya Expanse 8B weights and model release under Cohere Labs' CC-BY-NC terms", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-15" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions referenced by the official Aya Expanse model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-04" } ], "access_boundary": "This record represents the Aya Expanse 8B research release. Cohere's managed API, downstream fine-tunes, inference hosts, and larger Aya variants are separate records or service boundaries.", "strengths": [ "Developer-published multilingual weights with a detailed first-party model card", "The governing noncommercial and acceptable-use restrictions are explicit" ], "tradeoffs": [ "The noncommercial boundary prevents general production reuse without separate permission", "The source metadata does not establish broad runtime or serving coverage" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.62 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/CohereLabs/aya-expanse-8b", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/ef71a4865a5821d8cd24f171a0100156a02e808e/models/cohere/c4ai-aya-expanse-8b.toml", "verified_at": "2026-09-04" } ], "metadata_verified_at": "2026-09-04", "verified_at": "2026-09-04" }, { "id": "model-cohere-c4ai-aya-vision-32b", "source_id": "cohere/c4ai-aya-vision-32b", "name": "Aya Vision 32B", "developer": "Cohere Labs", "url": "https://huggingface.co/CohereLabs/aya-vision-32b", "description": "Cohere Labs' 32-billion-parameter multilingual vision-language release accepting image and text input, published as gated weights under CC-BY-NC-4.0 with a hosted Chat API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Aya Vision 32B", "description": "Open multilingual vision model for OCR, visual reasoning, and image question answering", "family": null, "release_date": "2025-03-04", "last_updated": "2025-05-14", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 16000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": "CC-BY-NC-4.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-vision-32b" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs aya-vision-32b weights and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/aya-vision-32b weight release of March 2025 with image and text input and text output, together with the hosted c4ai-aya-vision-32b Chat API model. The 8B sibling, the Expanse text lines, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 32B vision-language weights under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Cohere's docs serve c4ai-aya-vision-32b on the Chat API with image input" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "Self-operation needs a non-stock transformers build for the vision adapter" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.56 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/aya-vision-32b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Cohere Aya Vision documentation serving c4ai-aya-vision-32b", "url": "https://docs.cohere.com/docs/aya-vision", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/c4ai-aya-vision-32b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-c4ai-aya-vision-8b", "source_id": "cohere/c4ai-aya-vision-8b", "name": "Aya Vision 8B", "developer": "Cohere Labs", "url": "https://huggingface.co/CohereLabs/aya-vision-8b", "description": "Cohere Labs' 8-billion-parameter multilingual vision-language release accepting image and text input, published as gated weights under CC-BY-NC-4.0 without a managed API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Aya Vision 8B", "description": "Compact open multilingual vision model for OCR and visual question answering", "family": null, "release_date": "2025-03-04", "last_updated": "2025-05-14", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 16000, "input": null, "output": 4000 }, "reported_open_weights": true, "reported_license": "CC-BY-NC-4.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-vision-8b" } ] }, "licenses": [ "CC-BY-NC-4.0", "LicenseRef-Cohere-Labs-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The official model card releases these weights under CC-BY-NC-4.0 and requires compliance with the Cohere Labs Acceptable Use Policy, which the license incorporates by reference. The noncommercial limit and the use restrictions mean the public weights are not an open-source distribution; commercial use goes through separate Cohere API terms.", "license_evidence": [ { "license_id": "CC-BY-NC-4.0", "scope": "CohereLabs aya-vision-8b weights and model release under Cohere's CC-BY-NC 4.0 license with acceptable-use addendum", "kind": "web_terms", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-Cohere-Labs-AUP", "scope": "Mandatory acceptable-use restrictions incorporated by reference into the CC-BY-NC license and named on the official model card", "kind": "web_terms", "url": "https://docs.cohere.com/docs/cohere-labs-acceptable-use-policy", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the CohereLabs/aya-vision-8b weight release of March 2025 with image and text input and text output. The 32B sibling, the Expanse text lines, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Gated 8B vision-language weights under explicit CC-BY-NC-4.0 plus acceptable-use terms", "Vision adapter over the multilingual Expanse recipe with a 16K context window" ], "tradeoffs": [ "Gated access and noncommercial-only weights keep this out of open-source distribution", "No managed API serves this exact ID; use is self-operated or demo-only" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.28 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/CohereLabs/aya-vision-8b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/c4ai-aya-vision-8b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-cohere-north-mini-code-1-0", "source_id": "cohere/north-mini-code-1-0", "name": "North Mini Code", "developer": "Cohere Labs", "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", "description": "Cohere Labs' 30-billion-parameter Mixture-of-Experts coding release with 3B active parameters, published as ungated Apache-2.0 weights with managed API and Model Vault serving.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "North Mini Code", "description": "Cohere coding model for practical software engineering and agentic edits", "family": "north", "release_date": "2026-06-09", "last_updated": "2026-06-09", "knowledge_cutoff": "2025-09-23", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 64000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card and docs page declare North Mini Code released under Apache 2.0 and suitable for local deployment, with no acceptable-use addendum or noncommercial cap, unlike the CC-BY-NC Command lines. No LICENSE file ships in the weight repository, so the grant rests on those concordant first-party declarations pointing at the standard Apache-2.0 text.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "North Mini Code weights as declared on the official model card and docs page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the June 2026 North-Mini-Code-1.0 release, the inaugural member of Cohere's North family of code models, together with its Chat API and Model Vault serving. The North enterprise platform, later North family members, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated 30B MoE coding weights under a declared Apache-2.0 grant", "First-party Chat API and Model Vault production serving for the exact model ID" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card and docs declarations", "A three-month-old line with no retirement posture yet" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.74 }, "evidence": [ { "kind": "web", "label": "Official model card declaring the Apache-2.0 grant", "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official North Mini Code docs page declaring the Apache-2.0 grant", "url": "https://docs.cohere.com/docs/north-mini-code-1.0", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/cohere/north-mini-code-1-0.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-deepreinforce-ornith-1-0-35b", "source_id": "deepreinforce/ornith-1.0-35b", "name": "Ornith 1.0 35B", "developer": "DeepReinforce", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B", "description": "DeepReinforce's Ornith 1.0 mid-size release, published as MIT-claimed open weights without a license file.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ornith 1.0 35B", "description": "Large coding-reasoning model for agentic software tasks and RL search", "family": "ornith", "release_date": "2026-06-25", "last_updated": "2026-06-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MIT", "links": [ { "label": "Model card", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B", "type": "model_card" }, { "label": "Announcement", "url": "https://deep-reinforce.com/ornith_1_0.html", "type": "announcement" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card frontmatter claims the MIT license with a license link, but no LICENSE file ships in the weight repository, so the grant rests on that declaration pointing at the standard MIT text. No use restrictions accompany the claim.", "license_evidence": [ { "license_id": "MIT", "scope": "Ornith-1.0-35B weights as declared on the official model card, pointing at the standard MIT text", "kind": "web_terms", "url": "https://opensource.org/licenses/MIT", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the ornith-ai/Ornith-1.0-35B weight release of June 2026 at Hugging Face commit 5df2ed3f675c7beaa490328cc70bb573b65fb660. Other Ornith generations and community quantizations remain separate boundaries.", "strengths": [ "Ungated mid-size checkpoint under a declared MIT grant", "Dated June 2026 release with pinned commit history" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on frontmatter declarations", "No managed API serves this line anywhere" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 7.32 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepreinforce/ornith-1.0-35b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-deepreinforce-ornith-1-0-397b", "source_id": "deepreinforce/ornith-1.0-397b", "name": "Ornith 1.0 397B", "developer": "DeepReinforce", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B", "description": "DeepReinforce's Ornith 1.0 flagship release, published as MIT-claimed open weights without a license file.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ornith 1.0 397B", "description": "Large coding-reasoning model for agentic software tasks and RL search", "family": "ornith", "release_date": "2026-06-25", "last_updated": "2026-06-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MIT", "links": [ { "label": "Model card", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B", "type": "model_card" }, { "label": "Announcement", "url": "https://deep-reinforce.com/ornith_1_0.html", "type": "announcement" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { "label": "Hugging Face (FP8)", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B-FP8", "quantization": "fp8" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card frontmatter claims the MIT license with a license link, but no LICENSE file ships in the weight repository, so the grant rests on that declaration pointing at the standard MIT text. No use restrictions accompany the claim.", "license_evidence": [ { "license_id": "MIT", "scope": "Ornith-1.0-397B weights as declared on the official model card, pointing at the standard MIT text", "kind": "web_terms", "url": "https://opensource.org/licenses/MIT", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the ornith-ai/Ornith-1.0-397B weight release of June 2026 at Hugging Face commit 5e3e761811e804c295c1d3c0ce68b21da6154209. Other Ornith generations and community quantizations remain separate boundaries.", "strengths": [ "Ungated flagship checkpoint under a declared MIT grant", "Dated June 2026 release with pinned commit history" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on frontmatter declarations", "397B total parameters need cluster-class serving with no managed option" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 5, "serving_reach": 4, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 6.72 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepreinforce/ornith-1.0-397b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-deepreinforce-ornith-1-0-9b", "source_id": "deepreinforce/ornith-1.0-9b", "name": "Ornith 1.0 9B", "developer": "DeepReinforce", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B", "description": "DeepReinforce's Ornith 1.0 small release, published as MIT-claimed open weights without a license file.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ornith 1.0 9B", "description": "Open coding-reasoning model for repository tasks and self-improving agents", "family": "ornith", "release_date": "2026-06-25", "last_updated": "2026-06-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MIT", "links": [ { "label": "Model card", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B", "type": "model_card" }, { "label": "Announcement", "url": "https://deep-reinforce.com/ornith_1_0.html", "type": "announcement" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card frontmatter claims the MIT license with a license link, but no LICENSE file ships in the weight repository, so the grant rests on that declaration pointing at the standard MIT text. No use restrictions accompany the claim.", "license_evidence": [ { "license_id": "MIT", "scope": "Ornith-1.0-9B weights as declared on the official model card, pointing at the standard MIT text", "kind": "web_terms", "url": "https://opensource.org/licenses/MIT", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the ornith-ai/Ornith-1.0-9B weight release of June 2026 at Hugging Face commit 83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999. Other Ornith generations and community quantizations remain separate boundaries.", "strengths": [ "Ungated small checkpoint under a declared MIT grant, lightest in its generation", "Dated June 2026 release with pinned commit history" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on frontmatter declarations", "No managed API serves this line anywhere" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 7.32 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepreinforce/ornith-1.0-9b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-deepreinforce-ornith-1-5-35b-a3b", "source_id": "deepreinforce/ornith-1.5-35b-a3b", "name": "Ornith 1.5 35B A3B", "developer": "DeepReinforce", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", "description": "DeepReinforce's Ornith 1.5 mid-size release extending self-improvement to task generation, published as MIT-claimed open weights without a license file.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ornith 1.5 35B A3B", "description": "Mixture-of-experts coding-reasoning model for agentic software tasks, tool use, and image understanding", "family": "ornith", "release_date": "2026-08-18", "last_updated": "2026-08-23", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MIT", "links": [ { "label": "Model card", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", "type": "model_card" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card frontmatter claims the MIT license with a license link, but no LICENSE file ships in the weight repository, so the grant rests on that declaration pointing at the standard MIT text. No use restrictions accompany the claim.", "license_evidence": [ { "license_id": "MIT", "scope": "Ornith-1.5-35B-A3B weights as declared on the official model card, pointing at the standard MIT text", "kind": "web_terms", "url": "https://opensource.org/licenses/MIT", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the ornith-ai/Ornith-1.5-35B-A3B weight release of August 2026 at Hugging Face commit 10fbf86fed7ecee4a061f8b499a618f46001cac1, a new generation rather than a 1.0 snapshot. Other Ornith generations and community quantizations remain separate boundaries.", "strengths": [ "Ungated mid-size checkpoint under a declared MIT grant with GGUF, MLX, and NVFP4 variants", "Dated August 2026 release with dedicated methods announcement" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on frontmatter declarations", "No managed API serves this line anywhere" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 7.32 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepreinforce/ornith-1.5-35b-a3b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-deepseek-deepseek-ocr-2", "source_id": "deepseek/deepseek-ocr-2", "name": "DeepSeek OCR 2", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2", "description": "DeepSeek's 3-billion-parameter OCR release mapping document images to markdown, published as Apache-2.0 weights with no managed API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "DeepSeek OCR 2", "description": "High-accuracy OCR model for extracting text from documents, screenshots, receipts, and natural scenes", "family": null, "release_date": "2026-01-27", "last_updated": "2026-01-27", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": null }, "limits": { "context": 8192, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.txt file is the verbatim Apache License 2.0 text, appendix completed as Copyright 2023 DeepSeek, with no use restrictions attached. No separate model-license gate was found, so the release is open source.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "DeepSeek-OCR-2 model weights and code in the official repository", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/LICENSE.txt", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the deepseek-ai/DeepSeek-OCR-2 weight release of January 2026 at Hugging Face commit aaa02f3811945a91062062994c5c4a3f4c0af2b0 (last modified 2026-02-03), with image and text input and text output. Other DeepSeek lines and community quantizations remain separate boundaries.", "strengths": [ "Ungated 3B checkpoint carrying a verbatim Apache-2.0 license file beside the weights", "Single-file weights with tokenizer and modeling code for document OCR" ], "tradeoffs": [ "No managed API serves this release and no inference provider deploys it", "No output-limit statement accompanies the 8K native context" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.86 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepseek/deepseek-ocr-2.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-deepseek-deepseek-v3-2", "source_id": "deepseek/deepseek-v3.2", "name": "DeepSeek V3.2", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "description": "A developer-published hybrid-reasoning language-model release whose downloadable weights and repository are licensed under the MIT License.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "DeepSeek V3.2", "description": "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes, sparse attention, and tool-use", "family": "deepseek", "release_date": "2025-12-01", "last_updated": "2025-12-01", "knowledge_cutoff": "2024-07", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 64000 }, "reported_open_weights": true, "reported_license": "MIT License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified MIT License, and the model card states that the repository and the model weights are licensed under it. No separate model license or acceptable-use policy is incorporated.", "license_evidence": [ { "license_id": "MIT", "scope": "deepseek-ai/DeepSeek-V3.2 repository code, encoding utilities, and model weights", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the deepseek-ai/DeepSeek-V3.2 repository release at commit a7e62ac04ecb2c0a54d736dc46601c5606cf10a6 (last modified 2025-12-01). DeepSeek-V3.2-Speciale, DeepSeek-V3.2-Exp and its base model, the DeepSeek API's deepseek-chat and deepseek-reasoner names, and downstream quantizations remain separate boundaries.", "strengths": [ "Developer-published Safetensors shards, tokenizer, encoding utilities, and a repository MIT license file that the model card applies to the weights", "The model card distinguishes this release from the Speciale and Exp variants, documents the thinking-with-tools message format, and links a technical report" ], "tradeoffs": [ "Self-hosting guidance is delegated to the DeepSeek-V3.2-Exp repository rather than documented in this release's own card", "DeepSeek's API change log retired the model names that served this release, so first-party managed access has ended and the card itself carries no dates" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.48 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official DeepSeek API change log documenting the V3.2 rollout and retirement of its API model names", "url": "https://api-docs.deepseek.com/updates", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/deepseek/deepseek-v3.2.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-deepseek-deepseek-v4-flash-0423", "source_id": "deepseek/deepseek-v4-flash-0423", "name": "DeepSeek V4 Flash 0423", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "description": "DeepSeek's official April 23, 2026 preview snapshot of the V4 Flash line, a 284-billion-parameter MoE release with 1M-token context and built-in speculative decoding, published under the standard DeepSeek MIT license.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "DeepSeek V4 Flash 0423", "description": "Initial DeepSeek V4 Flash snapshot for economical reasoning, coding, and million-token agent workloads", "family": "deepseek-flash", "release_date": "2026-04-23", "last_updated": "2026-04-23", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 384000 }, "reported_open_weights": true, "reported_license": "MIT", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The DeepSeek-V4-Flash repository carries the standard DeepSeek MIT license file at the repository root, and the model card states the repository and weights are MIT-licensed. The dated 0423 identity comes from models.dev's snapshot naming; DeepSeek's own name for this release is the V4 Flash preview.", "license_evidence": [ { "license_id": "MIT", "scope": "Official DeepSeek-V4-Flash repository, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the dated April 23, 2026 snapshot of the DeepSeek V4 Flash line, published as DeepSeek-V4-Flash in the deepseek-ai organization under the standard DeepSeek MIT license. The undated deepseek-v4-flash API line, the July 31 dated snapshot, the V4.1 successor line, and independent hosts are separate boundaries.", "strengths": [ "DeepSeek publishes ungated Safetensors weights with a pinned MIT license file in the repository root", "vLLM and SGLang serving recipes are documented with the release" ], "tradeoffs": [ "The dated identity rests on models.dev's snapshot naming; DeepSeek's own naming is the V4 Flash preview without a literal 0423 tag", "The first-party deepseek-v4-flash API name is retired and routes to V4.1 Flash, so no current managed access to these weights" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 8.16 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepseek/deepseek-v4-flash-0423.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-deepseek-deepseek-v4-flash-0731", "source_id": "deepseek/deepseek-v4-flash-0731", "name": "DeepSeek V4 Flash 0731", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "description": "DeepSeek's official 304-billion-parameter Flash release with 1M-token context and built-in speculative decoding for coding and long agent runs, published under the MIT License with weights on Hugging Face and exact-model third-party hosting.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "DeepSeek V4 Flash 0731", "description": "Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding", "family": "deepseek-flash", "release_date": "2026-07-31", "last_updated": "2026-07-31", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 384000 }, "reported_open_weights": true, "reported_license": "MIT", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the verbatim MIT License, and the model card states that the repository and the model weights are licensed under it, with no additional model-license gates. NVIDIA's host documentation independently records the weights terms as MIT. First-party API consumption was governed separately by DeepSeek Open Platform terms while served and does not alter the weights license.", "license_evidence": [ { "license_id": "MIT", "scope": "Repository and model weights in deepseek-ai/DeepSeek-V4-Flash-0731; API consumption governed separately by DeepSeek Open Platform terms while served", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/main/LICENSE", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the DeepSeek-V4-Flash-0731 official release checkpoint (304B total, 1M context, speculative-decoding module attached) released July 31, 2026. The V4 Flash preview sibling, the V4 Pro siblings and dated snapshots, the experimental multimodal Vision sibling, the V4.1 Flash successor line that the deepseek-v4-flash API name routes to, and community quantizations and fine-tunes remain separate boundaries.", "strengths": [ "MIT-licensed 304B weights with encoding folder, inference scripts, and documented vLLM and SGLang speculative-decoding recipes", "Exact-model NVIDIA NIM hosting with first-party reference and catalog documentation" ], "tradeoffs": [ "First-party DeepSeek API serving has ended: the deepseek-v4-flash name is retired and routed to V4.1 Flash, so managed access no longer serves these weights", "No Jinja chat template in-repo; message encoding requires the dedicated encoding-folder scripts" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.42 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "verified_at": "2026-09-12" }, { "kind": "web", "label": "DeepSeek API change log: July 2026 official release and September 2026 retirement routing", "url": "https://api-docs.deepseek.com/updates", "verified_at": "2026-09-12" }, { "kind": "web", "label": "NVIDIA NIM reference for deepseek-ai/deepseek-v4-flash-0731", "url": "https://docs.api.nvidia.com/nim/reference/deepseek-ai-deepseek-v4-flash-0731", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/deepseek/deepseek-v4-flash-0731.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12" }, { "id": "model-deepseek-deepseek-v4-flash-vision-exp", "source_id": "deepseek/deepseek-v4-flash-vision-exp", "name": "DeepSeek V4 Flash Vision Exp", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp", "description": "DeepSeek's 305-billion-parameter experimental multimodal release with image input, published as MIT weights with a retired managed API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "DeepSeek V4 Flash Vision Exp", "description": "Experimental multimodal DeepSeek V4 Flash model for image understanding, coding, and agentic work", "family": "deepseek-flash", "release_date": "2026-08-21", "last_updated": "2026-08-21", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 384000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the verbatim MIT text, appendix completed as Copyright 2023 DeepSeek, with no use restrictions attached, so the release is open source.", "license_evidence": [ { "license_id": "MIT", "scope": "DeepSeek-V4-Flash-Vision-Exp model weights under the repository MIT license", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the deepseek-ai/DeepSeek-V4-Flash-Vision-Exp weight release of August 2026 at Hugging Face commit 6821d6ad3681a4b137b066b76094fa82ebd0a380 (last modified 2026-09-01), a genuinely new multimodal line rather than a snapshot of the reviewed text-only Flash-0731 record. Its deepseek-v4-flash-vision-exp API name retired on September 10, 2026 routing to V4.1 Flash. The Flash-0731 line, other V4 lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 305B multimodal checkpoint carrying a verbatim MIT license file beside the weights", "Dated August 2026 release with explicitly documented retirement and routing" ], "tradeoffs": [ "The managed API retired after three weeks, leaving self-operation as the supported path", "The Exp tag leaves the support window unclear" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.38 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepseek/deepseek-v4-flash-vision-exp.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-deepseek-deepseek-v4-pro", "source_id": "deepseek/deepseek-v4-pro", "name": "DeepSeek V4 Pro", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813", "description": "DeepSeek's 1.65-trillion-parameter open MoE flagship with 1M-token context for coding and long agent runs, published under the MIT License with weights on Hugging Face and served via the DeepSeek API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "DeepSeek V4 Pro", "description": "Open MoE flagship with million-token context for coding and long agent runs", "family": "deepseek-thinking", "release_date": "2026-04-24", "last_updated": "2026-04-24", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 384000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the verbatim MIT License, and the model card states that the repository and the model weights are licensed under it, with no additional model-license gates. API consumption is governed separately by the DeepSeek Open Platform terms, which do not alter the weights license. Unlike DeepSeek-V3, this release does not use the DeepSeek Model License 1.0.", "license_evidence": [ { "license_id": "MIT", "scope": "Repository and model weights in deepseek-ai/DeepSeek-V4-Pro-0813; API consumption governed separately by DeepSeek Open Platform terms", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/main/LICENSE", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the DeepSeek-V4-Pro-0813 official release checkpoint (deepseek-ai/DeepSeek-V4-Pro-0813, 1.65T total, 49B active, 1M context, speculative-decoding module attached) and its deepseek-v4-pro API model ID, which serves this release with no migration; DeepSeek reversed the planned September routing of the ID to V4.1 Flash and continues serving V4 Pro. The preview checkpoint repository, the Base sibling, the dated 0423 snapshot, the V4 Flash siblings, the V4.1 Flash successor line, and community quantizations and fine-tunes remain separate boundaries.", "strengths": [ "MIT-licensed 1.65T official-release weights with mirrors and standard Transformers, vLLM, SGLang, and NIM paths", "First-party DeepSeek API with OpenAI- and Anthropic-compatible endpoints, 1M context, and 384K maximum output plus documented NVIDIA NIM hosting for the exact checkpoint" ], "tradeoffs": [ "Checkpoint scale keeps single-node self-hosting heavy despite efficient hybrid attention", "The preview-to-official transition replaced the weights checkpoint and dropped the Jinja chat template for encoding-folder scripts" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.56 }, "evidence": [ { "kind": "web", "label": "Official DeepSeek-V4-Pro-0813 weights repository and model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813", "verified_at": "2026-09-12" }, { "kind": "web", "label": "MIT LICENSE blob for the 0813 release", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/main/LICENSE", "verified_at": "2026-09-12" }, { "kind": "web", "label": "DeepSeek API change log: official release and continued V4 Pro serving", "url": "https://api-docs.deepseek.com/updates", "verified_at": "2026-09-12" }, { "kind": "web", "label": "DeepSeek API pricing showing the 0813 model version", "url": "https://api-docs.deepseek.com/quick_start/pricing", "verified_at": "2026-09-12" }, { "kind": "web", "label": "NVIDIA NIM model page for deepseek-v4-pro-0813", "url": "https://build.nvidia.com/deepseek-ai/deepseek-v4-pro-0813", "verified_at": "2026-09-15" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/deepseek/deepseek-v4-pro.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-12" }, { "id": "model-deepseek-deepseek-v4-1-flash", "source_id": "deepseek/deepseek-v4.1-flash", "name": "DeepSeek V4.1 Flash", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "description": "DeepSeek's September 10, 2026 V4.1 Flash release, a text-and-image MoE with 1M-token context published as MIT open weights and served as the canonical deepseek-flash API line.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "DeepSeek V4.1 Flash", "description": "DeepSeek V4.1 Flash model for reasoning and agentic coding", "family": "deepseek-flash", "release_date": "2026-09-10", "last_updated": "2026-09-10", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 384000 }, "reported_open_weights": true, "reported_license": "MIT", "links": [], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The DeepSeek-V4.1-Flash repository carries the standard DeepSeek MIT license file at the repository root, and the model card states the repository and weights are MIT-licensed.", "license_evidence": [ { "license_id": "MIT", "scope": "Official DeepSeek-V4.1-Flash repository, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the DeepSeek-V4.1-Flash release of September 10, 2026, a distinct line in DeepSeek's newer CED architecture family published in the deepseek-ai organization under the standard DeepSeek MIT license. The V4 Flash preview snapshots, the retired deepseek-v4-flash API alias, and independent hosts are separate boundaries.", "strengths": [ "Ungated MIT-licensed weights with a pinned LICENSE file at the repository root", "Live on the DeepSeek API as the canonical deepseek-flash model and on several independent hosts" ], "tradeoffs": [ "In-day revision churn on the release commit leaves the pinned snapshot slightly ambiguous", "The model card cites a 552B backbone while the repository lists 763B of safetensors, so parameter counts need care" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 9, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 8.56 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepseek/deepseek-v4.1-flash.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-deepseek-deepseek-r1", "source_id": "deepseek/deepseek-r1", "name": "DeepSeek-R1", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1", "description": "A developer-published reasoning language-model release whose downloadable weights and repository code are licensed under the MIT License.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "DeepSeek-R1", "description": "Classic open reasoning model for transparent math, coding, and deliberate problem solving", "family": "deepseek-thinking", "release_date": "2025-01-20", "last_updated": "2025-05-29", "knowledge_cutoff": "2024-07", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified MIT License, and the model card states that both the code repository and the model weights are licensed under it, with commercial use, modification, and distillation expressly permitted. The distilled companion models carry the upstream Qwen and Llama licenses and are outside this record.", "license_evidence": [ { "license_id": "MIT", "scope": "deepseek-ai/DeepSeek-R1 repository code and model weights", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the deepseek-ai/DeepSeek-R1 repository release at commit 56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad (last modified 2025-03-27). DeepSeek-R1-Zero, the six distilled models, the separate DeepSeek-R1-0528 repository, downstream quantizations, and the DeepSeek API remain separate boundaries.", "strengths": [ "Developer-published Safetensors shards, tokenizer, and modeling code with a repository MIT license file that the model card applies to the weights", "The model card distinguishes DeepSeek-R1 from DeepSeek-R1-Zero and the distilled variants and names vLLM and SGLang as self-operated serving paths" ], "tradeoffs": [ "The model card's self-hosting examples target the distilled models, and the very large full checkpoint narrows practical self-operation", "The 0528 update is a separate repository and DeepSeek's API change log shows this release is no longer served first-party, so lifecycle facts live outside the model card" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.36 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official DeepSeek API change log documenting which releases the API has served", "url": "https://api-docs.deepseek.com/updates", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/deepseek/deepseek-r1.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-deepseek-deepseek-r1-0528", "source_id": null, "name": "DeepSeek-R1-0528", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "description": "A developer-published May 2025 reasoning language-model update whose downloadable weights and repository code are licensed under the MIT License.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "DeepSeek-R1-0528", "description": "Minor version upgrade of DeepSeek R1 with deeper reasoning, reduced hallucination, and function calling support", "family": "deepseek-thinking", "release_date": "2025-05-28", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 163840, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified MIT License, and the model card states that both the code repository and the use of DeepSeek-R1 models are under it, with commercial use and distillation expressly supported. The distilled DeepSeek-R1-0528-Qwen3-8B companion carries its upstream Qwen terms and is outside this record.", "license_evidence": [ { "license_id": "MIT", "scope": "deepseek-ai/DeepSeek-R1-0528 repository code and model weights", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/blob/4236a6af538feda4548eca9ab308586007567f52/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the deepseek-ai/DeepSeek-R1-0528 repository release at commit 4236a6af538feda4548eca9ab308586007567f52 (last modified 2025-05-29). The original DeepSeek-R1 repository release, DeepSeek-R1-0528-Qwen3-8B, downstream quantizations, and the DeepSeek API remain separate boundaries.", "strengths": [ "Developer-published Safetensors shards, tokenizer, and modeling code with an MIT license file that the model card applies to the weights", "DeepSeek's API change log dates when deepseek-reasoner moved to this release and when it moved on to V3.1" ], "tradeoffs": [ "The very large full checkpoint narrows practical self-operation, and the card defers local-run instructions to a separate repository", "DeepSeek's API change log shows this release is no longer served first-party, and the card carries no explicit release date" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.36 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official DeepSeek API change log documenting the 0528 upgrade and later API upgrades", "url": "https://api-docs.deepseek.com/updates", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-deepseek-deepseek-r1-distill-qwen-32b", "source_id": "deepseek/deepseek-r1-distill-qwen-32b", "name": "DeepSeek-R1-Distill-Qwen-32B", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", "description": "DeepSeek's 32-billion-parameter R1 reasoning distill of Qwen2.5-32B, published as MIT weights with Qwen Apache provenance and no managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "DeepSeek-R1-Distill-Qwen-32B", "description": "R1 reasoning distilled into Qwen 2.5 32B for efficient open-weight step-by-step problem solving", "family": "deepseek-thinking", "release_date": "2025-01-20", "last_updated": "2025-01-20", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B" } ] }, "licenses": [ "MIT", "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the MIT text stating the model weights are licensed under it; the card discloses the Qwen2.5 Apache-2.0 upstream the distill was finetuned from. Both grants are permissive with no use restrictions, so the release is open source with both provenances recorded.", "license_evidence": [ { "license_id": "MIT", "scope": "DeepSeek-R1-Distill-Qwen-32B model weights under the repository MIT license", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/blob/711ad2ea6aa40cfca18895e8aca02ab92df1a746/LICENSE", "verified_at": "2026-09-18" }, { "license_id": "Apache-2.0", "scope": "Qwen2.5 upstream provenance disclosed on the official model card", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the deepseek-ai/DeepSeek-R1-Distill-Qwen-32B weight release of January 2025 at Hugging Face commit 711ad2ea6aa40cfca18895e8aca02ab92df1a746 (last modified 2025-02-24). The full R1 line served under deepseek-reasoner, sibling distills of other sizes, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 32B reasoning distill with explicit MIT weights scope plus disclosed Qwen provenance", "Standard dense architecture deployable like Qwen and Llama models" ], "tradeoffs": [ "No managed API ever served the distill lines; only the full R1 reached the API", "Single release commit with the superseding R1-0528 lineage living elsewhere" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 8, "overall": 7.62 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/deepseek/deepseek-r1-distill-qwen-32b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-deepseek-deepseek-v3", "source_id": "deepseek/deepseek-v3", "name": "DeepSeek-V3", "developer": "DeepSeek", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3", "description": "A developer-published mixture-of-experts language-model release whose downloadable weights are governed by the DeepSeek Model License rather than an OSI-approved software license.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "DeepSeek-V3", "description": "Open DeepSeek MoE chat model for coding, math, and general reasoning", "family": "deepseek", "release_date": "2024-12-26", "last_updated": "2024-12-26", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "DeepSeek Model License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3" } ] }, "licenses": [ "LicenseRef-DeepSeek-Model-1.0" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The downloadable model weights use DeepSeek Model License 1.0, which grants broad reuse and commercial permissions while imposing use-based restrictions and downstream notice obligations.", "license_evidence": [ { "license_id": "LicenseRef-DeepSeek-Model-1.0", "scope": "DeepSeek-V3 Base and Chat model weights and derivatives", "kind": "web_terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3/blob/108e1e0b85635f3dd01561649ef21ad2be63749b/LICENSE-MODEL", "verified_at": "2026-09-04" } ], "access_boundary": "This record represents the DeepSeek-V3 provider-independent model release. The DeepSeek API, downstream quantizations and fine-tunes, and later V3 revisions remain separate boundaries.", "strengths": [ "Developer-published weight artifacts and model-specific governing terms", "The model card distinguishes the model license from the repository code license" ], "tradeoffs": [ "Use-based restrictions make the release source-available rather than open source", "The very large artifact narrows practical self-operation even though weights are downloadable" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.12 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/ef71a4865a5821d8cd24f171a0100156a02e808e/models/deepseek/deepseek-v3.toml", "verified_at": "2026-09-04" } ], "metadata_verified_at": "2026-09-04", "verified_at": "2026-09-04" }, { "id": "model-google-deep-research-max-preview-04-2026", "source_id": "google/deep-research-max-preview-04-2026", "name": "Deep Research Max Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/deep-research-max-preview-04-2026", "description": "An agentic research release served through Google's Interactions API under Google's API terms, with a 1,048,576-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Deep Research Max Preview", "description": "Maximum-comprehensiveness agentic researcher for multi-step investigation, synthesis, and cited reports", "family": "gemini-pro", "release_date": "2026-04-21", "last_updated": "2026-04-21", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Deep Research Max agent preview identified by the dated code deep-research-max-preview-04-2026, launched April 21, 2026 and still served. The standard Deep Research tier, the Interactions API platform, and later Deep Research releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and lifecycle status", "The deprecations table and changelog name the release and its migration path" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Serving runs on the paid Gemini API tiers through a single interface surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.26 }, "evidence": [ { "kind": "web", "label": "Official Deep Research Max agent model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/deep-research-max-preview-04-2026", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Deep Research launch announcement", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/next-generation-gemini-deep-research/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/deep-research-max-preview-04-2026.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-diffusiongemma-26b-a4b-it", "source_id": "google/diffusiongemma-26b-a4b-it", "name": "DiffusionGemma 26B A4B IT", "developer": "Google", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "description": "Google DeepMind's ungated experimental open-weights multimodal model that generates text by discrete diffusion rather than token-by-token autoregression: an encoder-decoder over the 26B-total, 3.8B-active Gemma 4 mixture-of-experts architecture, denoising 256-token canvases with bidirectional attention for 256K-token context, released under Apache-2.0 and served on Vertex AI Model Garden.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "DiffusionGemma 26B A4B IT", "description": "Gemma 4 based discrete text diffusion model for fast chat, reasoning, and tool use", "family": "gemma", "release_date": "2026-06-09", "last_updated": "2026-07-15", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card links its license to Google's Gemma 4 license page, which carries the Apache License 2.0 text in full, and Google's own release blog states the weights are \"Released under an Apache 2.0 license\". The Gemma Terms of Use, whose appendix does not list DiffusionGemma, state that Gemma 4 terms are in the separate Gemma 4 license, so the Gemma Terms of Use use restrictions and the incorporated Gemma Prohibited Use Policy do not attach to this release. No LICENSE file ships in the repository, so the grant rests on the card metadata and the linked license page.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "scope": "Official google/diffusiongemma-26B-A4B-it model repository: Safetensors weights, tokenizer, processor, chat template, and scheduler configuration", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/README.md", "verified_at": "2026-10-01" }, { "license_id": "Apache-2.0", "kind": "web_terms", "scope": "Google's Gemma 4 license page, linked from the model card as the governing license text, carrying the Apache License 2.0 in full", "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "verified_at": "2026-10-01" } ], "access_boundary": "This record represents the google/diffusiongemma-26B-A4B-it weight release published on Hugging Face at commit f7f5b7f5fa82ffc52addd066915886d497f5517b, the instruction-tuned 26B-total, 3.8B-active DiffusionGemma checkpoint. The Gemma 4 26B A4B autoregressive base it is fine-tuned from, the Diffusion Explained sample application, the community FP8 and 4-bit quantizations and the 31 finetune derivatives, and the Vertex AI Model Garden deployment of the same weights are separate boundaries.", "strengths": [ "Ungated Apache-2.0 weights in an official Google Hugging Face organization, with the full Apache 2.0 text linked from the model card and the release blog", "First open-weights Google release built on discrete text diffusion rather than autoregression, documented end to end by a separate Diffusion Explained page covering Uniform State Diffusion, incremental prefill, and denoising", "First-party model card publishes the architecture table, 256-token canvas, 1024-token sliding window, 8-of-128 expert routing, recommended sampler settings, per-runtime deployment commands, a training-data description with a January 2025 cutoff, and a safety evaluation section", "Served from Google's own Vertex AI Model Garden, which the developer documentation links directly from the model overview" ], "tradeoffs": [ "Google documents the release as experimental, and its own comparison table shows it scoring below the Gemma 4 base it derives from on most text and vision benchmarks", "Google states the parallel-decoding advantage is aimed at low-concurrency single-user local use and diminishes under high-QPS cloud workloads, so the throughput case does not transfer to server deployments", "No LICENSE file ships in the repository, so the grant rests on the card metadata and the linked license page rather than an in-repo license artifact", "No deprecation or versioning posture is published for the line, and the repository's last commit is 15 July 2026" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.06 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/dbc891ced2fdbf01591e86345100e972b329b6a6/models/google/diffusiongemma-26b-a4b-it.toml", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official DiffusionGemma model card: architecture, benchmarks, sampler settings, training data, safety evaluation, and limitations", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Pinned model card at the reviewed repository commit, the license grant this record rests on", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/README.md", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Google's Gemma 4 license page carrying the Apache License 2.0 in full", "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official Gemma documentation: DiffusionGemma model overview, recommended serving configuration, and the Vertex AI Model Garden deployment path", "url": "https://ai.google.dev/gemma/docs/diffusiongemma", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official Diffusion Explained page: Uniform State Diffusion, incremental prefill and denoising, and the autoregression comparison", "url": "https://ai.google.dev/gemma/docs/diffusiongemma/explained", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Official release announcement: DiffusionGemma, 4x faster text generation on GPUs, released under Apache 2.0", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Pinned model configuration giving the 256-token canvas, DiffusionGemmaForBlockDiffusion architecture, and 30-layer sliding/full attention layout", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/config.json", "verified_at": "2026-10-01" }, { "kind": "web", "label": "Gemma Terms of Use, whose appendix excludes DiffusionGemma and which routes Gemma 4 terms to the separate Apache-2.0 license", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-10-01" } ], "metadata_verified_at": "2026-10-01", "verified_at": "2026-10-01" }, { "id": "model-google-gemini-2-0-flash", "source_id": "google/gemini-2.0-flash", "name": "Gemini 2.0 Flash", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash", "description": "A retired multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, shut down June 1, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.0 Flash", "description": "Earlier Gemini Flash workhorse for responsive multimodal apps and tool use", "family": "gemini-flash", "release_date": "2024-12-11", "last_updated": "2024-12-11", "knowledge_cutoff": "2024-06", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provided Gemini 2.0 Flash as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the Gemini 2.0 release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 2.0 Flash line identified by the model code gemini-2.0-flash (experimental December 2024, general availability February 2025, shut down June 1, 2026). The -001 and -exp versions, the Flash-Lite variant, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the full lifecycle from experimental preview to announced deprecation and executed shutdown with named replacements", "The model page, deprecations table, and changelog each name the release and its shutdown date" ], "tradeoffs": [ "No developer-published weights were ever available for self-operation", "The release is shut down, so new API traffic must migrate to the named successor" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini 2.0 Flash model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini 2.0 Flash launch announcement", "url": "https://developers.googleblog.com/en/the-next-chapter-of-the-gemini-era-for-developers/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-2.0-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-2-0-flash-lite", "source_id": "google/gemini-2.0-flash-lite", "name": "Gemini 2.0 Flash-Lite", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash-lite", "description": "A retired cost-efficient multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, shut down June 1, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.0 Flash-Lite", "description": "Low-latency Gemini model for high-volume multimodal and agent workloads", "family": "gemini-flash-lite", "release_date": "2024-12-11", "last_updated": "2024-12-11", "knowledge_cutoff": "2024-06", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provided Gemini 2.0 Flash as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the Gemini 2.0 release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 2.0 Flash-Lite line identified by the model code gemini-2.0-flash-lite (public preview February 2025, general availability February 25, 2025, shut down June 1, 2026). The -001 version, the 2.0 Flash variant, the retired Flash-Lite Preview, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the full lifecycle from experimental preview to announced deprecation and executed shutdown with named replacements", "The model page, deprecations table, and changelog each name the release and its shutdown date" ], "tradeoffs": [ "No developer-published weights were ever available for self-operation", "The release is shut down, so new API traffic must migrate to the named successor" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini 2.0 Flash-Lite model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash-lite", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini 2.0 Flash-Lite launch announcement", "url": "https://developers.googleblog.com/en/gemini-2-family-expands/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-2.0-flash-lite.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-2-5-computer-use-preview-10-2025", "source_id": "google/gemini-2.5-computer-use-preview-10-2025", "name": "Gemini 2.5 Computer Use Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-computer-use-preview-10-2025", "description": "An active computer-use agent preview served through Google's Gemini API under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.5 Computer Use Preview", "description": "Specialized Gemini 2.5 model for browser-control agents that automate UI tasks", "family": "gemini-pro", "release_date": "2025-10-07", "last_updated": "2025-10-07", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 64000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 2.5 Computer Use preview identified by the dated code gemini-2.5-computer-use-preview-10-2025, launched October 7, 2025 and still served with no shutdown date announced. The built-in computer use on Gemini 3 Pro and Flash, the Gemini 2.5 chat lines, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the dated preview identity, modalities, limits, and the agentic capability in a model page and guide", "The launch announcement and changelog name the release and its specialized use" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "As a legacy preview line it has no pinned stable date, only the preview deprecation policy" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Gemini 2.5 Computer Use model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-computer-use-preview-10-2025", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-2.5-computer-use-preview-10-2025.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-2-5-flash", "source_id": "google/gemini-2.5-flash", "name": "Gemini 2.5 Flash", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash", "description": "A pinned multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.5 Flash", "description": "Fast Gemini workhorse for multimodal apps where latency and price matter", "family": "gemini-flash", "release_date": "2025-06-17", "last_updated": "2025-06-17", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides Gemini 2.5 Flash as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-11", "scope": "Access to Gemini 2.5 Flash through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 2.5 Flash release identified by the model code gemini-2.5-flash (latest update June 2025). The retired gemini-2.5-flash-preview-09-2025 version, the Flash-Lite, Flash Image, Live, and TTS variants, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a stable model code, a latest-update date, and a stable/preview/experimental version naming scheme", "Google publishes a deprecations page that lists released and retired 2.5 Flash versions with earliest shutdown dates" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations policy lists earliest possible shutdown dates rather than a fixed notice period, so continued availability depends on Google's lifecycle decisions" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 2.5 Flash model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/google/gemini-2.5-flash.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-google-gemini-2-5-flash-image", "source_id": "google/gemini-2.5-flash-image", "name": "Gemini 2.5 Flash Image", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-image", "description": "An active but deprecated image generation and editing release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nano Banana", "description": "Nano Banana image model for fast generation, edits, and character-consistent assets", "family": "gemini-flash", "release_date": "2025-08-26", "last_updated": "2025-08-26", "knowledge_cutoff": "2024-06", "modalities": { "input": [ "text", "image" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 2.5 Flash Image (Nano Banana) stable release of October 2025 under the model code gemini-2.5-flash-image, deprecated with retirement extended to March 15, 2027. The shut-down preview endpoint, the Nano Banana Pro and Nano Banana 2 lines, the Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the model page, deprecation row, changelog, and migration guidance to Nano Banana 2 Lite", "Retirement was explicitly extended and re-targeted in the Agent Platform release notes" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Serving runs on the paid Gemini API tiers through a single interface surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Gemini 2.5 Flash Image model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-image", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official launch announcement", "url": "https://developers.googleblog.com/en/introducing-gemini-2-5-flash-image/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-2.5-flash-image.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-2-5-flash-tts", "source_id": "google/gemini-2.5-flash-tts", "name": "Gemini 2.5 Flash TTS", "developer": "Google", "url": "https://docs.cloud.google.com/text-to-speech/docs/gemini-tts", "description": "Google's generally available Gemini 2.5 Flash text-to-speech model, served through the Cloud Text-to-Speech API under Google Cloud's terms for single- and multi-speaker speech, with no downloadable model weights.", "model_type": "audio_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.5 Flash TTS", "description": "Speech generation model for controllable voice, narration, and audio delivery", "family": "gemini-flash", "release_date": "2025-09-30", "last_updated": "2025-12-10", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text" ], "output": [ "audio" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": 16384 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service of the Cloud Text-to-Speech API under the Google Cloud Platform Terms of Service. The terms bar copying, reverse engineering, and redistributing the services, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "Google Cloud Platform Terms of Service, the agreement that governs the Cloud Text-to-Speech API (restrictions in section 3.3)", "kind": "web_terms", "url": "https://cloud.google.com/terms", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Gemini 2.5 Flash TTS release identified by the Cloud Text-to-Speech model ID gemini-2.5-flash-tts, generally available since September 30, 2025. The Gemini API's separate gemini-2.5-flash-preview-tts preview code, the Gemini 2.5 Flash Lite TTS preview, the Gemini 3.1 and Gemini 3.8 TTS releases, and the Chirp voices remain separate model or service boundaries.", "strengths": [ "Google documents the model ID, text-in and audio-out data types, token limits, and speaker support on its Cloud Text-to-Speech page", "The Cloud Text-to-Speech release notes date the general-availability launch and name both model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Google's Cloud documentation fetched for this review gives no retirement date for the model, and the Gemini API lists its own preview-tts codes for Gemini 2.5 TTS with a November 17, 2026 earliest shutdown", "Cloud terms are a general customer agreement rather than a model-specific licence" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.88 }, "evidence": [ { "kind": "web", "label": "Cloud Text-to-Speech Gemini-TTS documentation listing gemini-2.5-flash-tts with its text input, audio output, and token limits", "url": "https://docs.cloud.google.com/text-to-speech/docs/gemini-tts", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Cloud Text-to-Speech release notes: general availability of gemini-2.5-flash-tts and gemini-2.5-pro-tts on September 30, 2025", "url": "https://docs.cloud.google.com/text-to-speech/docs/release-notes", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Google Cloud Platform Services Summary naming Text-to-Speech as a Google Cloud service", "url": "https://cloud.google.com/terms/services", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/google/gemini-2.5-flash-tts.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-gemini-2-5-flash-lite", "source_id": "google/gemini-2.5-flash-lite", "name": "Gemini 2.5 Flash-Lite", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-lite", "description": "An active multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.5 Flash-Lite", "description": "Lean Gemini 2.5 lane for cheap multimodal traffic and quick agents", "family": "gemini-flash-lite", "release_date": "2025-06-17", "last_updated": "2025-06-17", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 2.5 Flash-Lite release identified by the model code gemini-2.5-flash-lite (preview June 2025, general availability July 22, 2025, no shutdown date announced). The shut-down preview endpoint, the Flash sibling, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and a dated deprecations row", "The changelog and model index name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations page marks an earliest possible retirement date for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 2.5 Flash-Lite model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-lite", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-2.5-flash-lite.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-2-5-pro", "source_id": "google/gemini-2.5-pro", "name": "Gemini 2.5 Pro", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro", "description": "A pinned multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.5 Pro", "description": "Google's proven reasoning model for coding, math, and multimodal analysis", "family": "gemini-pro", "release_date": "2025-06-17", "last_updated": "2025-06-17", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides Gemini 2.5 Pro as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-11", "scope": "Access to Gemini 2.5 Pro through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 2.5 Pro release identified by the model code gemini-2.5-pro (latest update June 2025). The retired gemini-2.5-pro-preview-* versions, Gemini 2.5 Flash, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a stable model code, a latest-update date, and a stable/preview/experimental version naming scheme", "Google publishes a deprecations page that lists released and retired 2.5 Pro versions with earliest shutdown dates" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations policy lists earliest possible shutdown dates rather than a fixed notice period, so continued availability depends on Google's lifecycle decisions" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 2.5 Pro model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/google/gemini-2.5-pro.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-google-gemini-2-5-pro-tts", "source_id": "google/gemini-2.5-pro-tts", "name": "Gemini 2.5 Pro TTS", "developer": "Google", "url": "https://docs.cloud.google.com/text-to-speech/docs/gemini-tts", "description": "Google's generally available Gemini 2.5 Pro text-to-speech model, served through the Cloud Text-to-Speech API under Google Cloud's terms for single- and multi-speaker speech, with no downloadable model weights.", "model_type": "audio_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 2.5 Pro TTS", "description": "Speech generation model for controllable voice, narration, and audio delivery", "family": "gemini-pro", "release_date": "2025-09-30", "last_updated": "2025-12-10", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text" ], "output": [ "audio" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 32768, "input": null, "output": 16384 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service of the Cloud Text-to-Speech API under the Google Cloud Platform Terms of Service. The terms bar copying, reverse engineering, and redistributing the services, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "Google Cloud Platform Terms of Service, the agreement that governs the Cloud Text-to-Speech API (restrictions in section 3.3)", "kind": "web_terms", "url": "https://cloud.google.com/terms", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Gemini 2.5 Pro TTS release identified by the Cloud Text-to-Speech model ID gemini-2.5-pro-tts, generally available since September 30, 2025. The Gemini API's separate gemini-2.5-pro-preview-tts preview code, the Gemini 2.5 Flash Lite TTS preview, the Gemini 3.1 and Gemini 3.8 TTS releases, and the Chirp voices remain separate model or service boundaries.", "strengths": [ "Google documents the model ID, text-in and audio-out data types, token limits, and speaker support on its Cloud Text-to-Speech page", "The Cloud Text-to-Speech release notes date the general-availability launch and name both model IDs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Google's Cloud documentation fetched for this review gives no retirement date for the model, and the Gemini API lists its own preview-tts codes for Gemini 2.5 TTS with a November 17, 2026 earliest shutdown", "Cloud terms are a general customer agreement rather than a model-specific licence" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.88 }, "evidence": [ { "kind": "web", "label": "Cloud Text-to-Speech Gemini-TTS documentation listing gemini-2.5-pro-tts with its text input, audio output, and token limits", "url": "https://docs.cloud.google.com/text-to-speech/docs/gemini-tts", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Cloud Text-to-Speech release notes: general availability of gemini-2.5-flash-tts and gemini-2.5-pro-tts on September 30, 2025", "url": "https://docs.cloud.google.com/text-to-speech/docs/release-notes", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Google Cloud Platform Services Summary naming Text-to-Speech as a Google Cloud service", "url": "https://cloud.google.com/terms/services", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/google/gemini-2.5-pro-tts.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-gemini-3-flash-preview", "source_id": "google/gemini-3-flash-preview", "name": "Gemini 3 Flash Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-flash-preview", "description": "A preview multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights; the preview remains available while Google steers new production use toward stable successors.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3 Flash Preview", "description": "New Gemini flash lane bringing frontier-style multimodal reasoning to cheaper runs", "family": "gemini-flash", "release_date": "2025-12-17", "last_updated": "2025-12-17", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides Gemini 3 Flash Preview as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit reverse engineering, extracting, or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to Gemini 3 Flash Preview through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy; no weight license", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the preview release identified by the undated model code gemini-3-flash-preview (latest update December 2025; knowledge cutoff January 2025; inputs text, image, video, audio, PDF; output text). There is no dated snapshot behind the ID and no shutdown date announced; Google recommends gemini-3.5-flash or gemini-3.6-flash for new production use. The stable successors, Gemini 3 Pro previews, image-generation and audio variants, the Gemini consumer app, and the Vertex AI and AI Studio platforms themselves are separate service, application, or model boundaries.", "strengths": [ "Google documents a distinct model code, full modality matrix including computer use, token limits, update date, and knowledge cutoff", "The release is corroborated across the API changelog, the deprecations table, Vertex AI release notes, and a DeepMind model-cards entry" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The undated preview ID carries only a two-week deprecation guarantee with no committed shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3 Flash Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-flash-preview", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Gemini API deprecations table", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official release announcement", "url": "https://blog.google/products/gemini/gemini-3-flash", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/google/gemini-3-flash-preview.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-google-gemini-3-pro-preview", "source_id": "google/gemini-3-pro-preview", "name": "Gemini 3 Pro Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview", "description": "A preview multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights; Google shut the preview endpoint down on March 9, 2026 and directs users to Gemini 3.1 Pro Preview.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3 Pro Preview", "description": "Preview Gemini flagship for complex reasoning, coding, and rich multimodal prompts", "family": "gemini-pro", "release_date": "2025-11-18", "last_updated": "2025-11-18", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides Gemini 3 Pro Preview as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit reverse engineering, extracting, or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to Gemini 3 Pro Preview through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy; no weight license", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the preview release identified by the undated model code gemini-3-pro-preview (latest update November 2025; knowledge cutoff January 2025; inputs text, image, video, audio, PDF; output text). Google shut the preview endpoint down on March 9, 2026 and names gemini-3.1-pro-preview as the replacement; there is no dated snapshot behind the ID. The successor preview, Gemini 3 Flash Preview, the image-generation releases, the Gemini consumer app, and the Vertex AI and AI Studio platforms themselves are separate service, application, or model boundaries.", "strengths": [ "Google documents a distinct model code, full modality matrix, token limits, release date, and an explicit shutdown with a named replacement", "A DeepMind Gemini 3 Pro model card corroborates identity, multimodal scope, knowledge cutoff, and prohibited-use applicability" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The preview endpoint is shut down after a four-month service life, so the record documents a retired release and its replacement" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3 Pro Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Gemini API deprecations table", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Vertex AI Gemini 3 Pro page", "url": "https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-pro", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official release announcement", "url": "https://blog.google/products/gemini/gemini-3/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/google/gemini-3-pro-preview.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-google-gemini-3-1-flash-lite", "source_id": "google/gemini-3.1-flash-lite", "name": "Gemini 3.1 Flash Lite", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite", "description": "An active multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.1 Flash Lite", "description": "Low-latency Gemini model for high-volume multimodal and agent workloads", "family": "gemini-flash-lite", "release_date": "2026-05-07", "last_updated": "2026-05-07", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 3.1 Flash-Lite release identified by the model code gemini-3.1-flash-lite (preview March 2026, general availability May 7, 2026, earliest possible retirement May 7, 2027 with gemini-3.5-flash-lite named as replacement). The shut-down preview endpoint, the Flash sibling, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and a dated deprecations row", "The changelog and model index name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations page marks an earliest possible retirement date for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.1 Flash-Lite model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.1-flash-lite.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-1-flash-lite-preview", "source_id": "google/gemini-3.1-flash-lite-preview", "name": "Gemini 3.1 Flash Lite Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-preview", "description": "A retired multimodal Gemini preview release served through Google's Gemini API under Google's API terms, shut down May 25, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.1 Flash Lite Preview", "description": "Low-latency Gemini model for high-volume multimodal and agent workloads", "family": "gemini-flash-lite", "release_date": "2026-03-03", "last_updated": "2026-03-03", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3.1 Flash-Lite Preview identified by the model code gemini-3.1-flash-lite-preview, released March 3, 2026 and shut down May 25, 2026 with gemini-3.1-flash-lite as its replacement. The GA line, the Flash sibling, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the release, deprecation, and shutdown dates in the changelog and deprecations table", "The model page carries a shutdown banner naming the replacement" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is shut down and new traffic must migrate to the GA replacement" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.1 Flash-Lite Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API changelog documenting launch and shutdown", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.1-flash-lite-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-1-flash-live-preview", "source_id": "google/gemini-3.1-flash-live-preview", "name": "Gemini 3.1 Flash Live Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview", "description": "A legacy real-time dialogue preview release served through Google's Live API under Google's API terms, with text and audio output and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.1 Flash Live Preview", "description": "High-quality, low-latency Live API model for real-time dialogue and voice-first AI applications", "family": "gemini-flash", "release_date": "2026-03-26", "last_updated": "2026-03-26", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text", "audio" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3.1 Flash Live preview identified by the model code gemini-3.1-flash-live-preview, released March 2026 and still served with no committed shutdown date, marked legacy with gemini-3.8-live named as its successor. The Gemini 3.8 Live line, the translate and transcribe live lines, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and lifecycle status", "The changelog and deprecations page name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through the Live API surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.1 Flash Live Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.1-flash-live-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-1-flash-tts-preview", "source_id": "google/gemini-3.1-flash-tts-preview", "name": "Gemini 3.1 Flash TTS Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-tts-preview", "description": "A legacy Gemini 3.1 Flash text-to-speech preview that turns text into steerable single- or multi-speaker audio through the Gemini API and Cloud Text-to-Speech under Google's terms, with no downloadable model weights and an earliest shutdown of November 17, 2026.", "model_type": "audio_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.1 Flash TTS Preview", "description": "Low-latency speech generation with steerable prompts and expressive audio tags", "family": "gemini-flash", "release_date": "2026-04-15", "last_updated": "2026-04-15", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text" ], "output": [ "audio" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 8192, "input": null, "output": 16384 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Gemini 3.1 Flash TTS preview identified by the model code gemini-3.1-flash-tts-preview, launched April 15, 2026 per the changelog, marked legacy, and listed with an earliest shutdown date of November 17, 2026 and the Gemini 3.8 TTS models as replacements. The Gemini 3.8 Flash and Flash-Lite TTS releases, the Gemini 2.5 TTS lines, and the Gemini app products remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with the model code, text-in and audio-out data types, token limits, and a legacy notice naming its replacements", "The deprecations table lists an earliest shutdown date and the replacement models" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is a legacy preview with an earliest shutdown of November 17, 2026", "The deprecations table lists a February 26, 2026 release date that does not match the April 15, 2026 launch in the changelog" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.1 Flash TTS Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-tts-preview", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API changelog documenting the April 15, 2026 launch", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Cloud Text-to-Speech Gemini-TTS documentation listing this model id and its audio output", "url": "https://docs.cloud.google.com/text-to-speech/docs/gemini-tts", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/google/gemini-3.1-flash-tts-preview.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-gemini-3-1-pro-preview", "source_id": "google/gemini-3.1-pro-preview", "name": "Gemini 3.1 Pro Preview", "developer": "Google", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-1-pro", "description": "A preview multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights; active since February 2026 as the replacement for the shut-down Gemini 3 Pro Preview, with general availability still pending.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.1 Pro Preview", "description": "Reasoning-first Gemini preview for agentic coding and complex problem solving", "family": "gemini-pro", "release_date": "2026-02-19", "last_updated": "2026-02-19", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides Gemini 3.1 Pro Preview as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service. The terms permit customer applications but prohibit developing competing models and reverse engineering, extracting, or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to Gemini 3.1 Pro Preview through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service; no weight license", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the preview release identified by the model code gemini-3.1-pro-preview (latest update February 2026; inputs text, image, video, audio, PDF; output text), together with its customtools endpoint variant for custom-tool-heavy agentic workflows. No shutdown date is announced and no stable successor ID exists. The retired gemini-3-pro-preview predecessor, the stable Flash line, image and audio-only releases, the Gemini consumer app, and the Vertex AI and AI Studio platforms themselves are separate model, service, or application boundaries.", "strengths": [ "Google documents a distinct preview model code plus a customtools endpoint variant, full modality matrix, token limits, and release date", "The deprecations table and changelog record the predecessor shutdown and the alias switch in writing" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Seven months in preview with only a two-week deprecation guarantee and no committed shutdown or general-availability date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Enterprise Agent Platform Gemini 3.1 Pro page (model ID, modalities, limits, customtools variant)", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-1-pro", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official Gemini API deprecations table", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official launch announcement (February 19, 2026)", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/google/gemini-3.1-pro-preview.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12" }, { "id": "model-google-gemini-3-5-flash", "source_id": "google/gemini-3.5-flash", "name": "Gemini 3.5 Flash", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash", "description": "An active multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.5 Flash", "description": "Fast Gemini model balancing multimodal reasoning, tool use, and cost", "family": "gemini-flash", "release_date": "2026-05-19", "last_updated": "2026-05-19", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 3.5 Flash release identified by the model code gemini-3.5-flash (general availability May 19, 2026, no shutdown date announced), which became the model behind gemini-flash-latest. The lite tier, the 3.6 and 3.7 successors, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and a dated deprecations row", "The changelog and model index name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations page marks an earliest possible retirement date for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.5 Flash model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.5-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-5-flash-lite", "source_id": "google/gemini-3.5-flash-lite", "name": "Gemini 3.5 Flash Lite", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite", "description": "An active multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.5 Flash Lite", "description": "Fast Gemini model balancing multimodal reasoning, tool use, and cost", "family": "gemini-flash-lite", "release_date": "2026-07-21", "last_updated": "2026-07-21", "knowledge_cutoff": "2026-03", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 3.5 Flash-Lite release identified by the model code gemini-3.5-flash-lite (general availability July 21, 2026, no shutdown date announced), the named replacement for gemini-3.1-flash-lite. The flagship flash lines, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and a dated deprecations row", "The changelog and model index name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations page marks an earliest possible retirement date for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.5 Flash-Lite model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.5-flash-lite.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-5-live-translate-preview", "source_id": "google/gemini-3.5-live-translate-preview", "name": "Gemini 3.5 Live Translate Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-live-translate-preview", "description": "A real-time speech-to-speech translation preview release served through Google's Live API under Google's API terms, with audio and text transcript output and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.5 Live Translate Preview", "description": "Low-latency audio-to-audio model for real-time speech translation across 70+ languages", "family": "gemini-pro", "release_date": "2026-06-09", "last_updated": "2026-06-09", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "audio" ], "output": [ "audio", "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 131072, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3.5 Live Translate preview identified by the model code gemini-3.5-live-translate-preview, released June 9, 2026 and still served with no shutdown date announced. The Gemini 3.1 Flash Live and 3.5 Transcribe live lines, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and lifecycle status", "The changelog and deprecations page name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through the Live API surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.5 Live Translate Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-live-translate-preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.5-live-translate-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-5-transcribe-live", "source_id": "google/gemini-3.5-transcribe-live", "name": "Gemini 3.5 Transcribe Live", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-transcribe", "description": "A low-latency streaming speech-to-text release served through Google's Live API under Google's API terms, with text output and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.5 Transcribe Live", "description": "Speech transcription model for accurate audio-to-text and captioning workflows", "family": "gemini", "release_date": "2026-08-26", "last_updated": "2026-08-26", "knowledge_cutoff": null, "modalities": { "input": [ "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": null }, "limits": { "context": 0, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3.5 Transcribe Live variant of the Gemini 3.5 Transcribe line, generally available August 26, 2026 with no shutdown date announced and documented through the unary gemini-3.5-transcribe model page. The unary variant, the live translate and Flash Live lines, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the model page, GA changelog entry, and a dedicated live-transcription guide", "The published limit is a 10-minute audio duration per live session rather than a token context" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through the Live API surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.5 Transcribe model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-transcribe", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini Live API transcription guide", "url": "https://ai.google.dev/gemini-api/docs/live-api/live-transcribe", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.5-transcribe-live.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-6-flash", "source_id": "google/gemini-3.6-flash", "name": "Gemini 3.6 Flash", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash", "description": "An active multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.6 Flash", "description": "Fast Gemini model balancing multimodal reasoning, tool use, and cost", "family": "gemini-flash", "release_date": "2026-07-21", "last_updated": "2026-07-21", "knowledge_cutoff": "2026-03", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 3.6 Flash release identified by the model code gemini-3.6-flash (general availability July 21, 2026, no shutdown date announced), named as the recommended replacement for both gemini-2.0-flash and gemini-3-flash-preview. The lite tier, the 3.5 and 3.7 flash siblings, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and a dated deprecations row", "The changelog and model index name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations page marks an earliest possible retirement date for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.6 Flash model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.6-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-7-flash", "source_id": "google/gemini-3.7-flash", "name": "Gemini 3.7 Flash", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.7-flash", "description": "An active multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.7 Flash", "description": "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning", "family": "gemini-flash", "release_date": "2026-08-13", "last_updated": "2026-08-13", "knowledge_cutoff": "2026-03", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the stable Gemini 3.7 Flash release identified by the model code gemini-3.7-flash (general availability August 13, 2026, no shutdown date announced), the newest flagship of the flash line in this batch. The lite tier, the 3.5 and 3.6 flash predecessors, the Vertex AI and Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and a dated deprecations row", "The changelog and model index name the release and its family position" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations page marks an earliest possible retirement date for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.7 Flash model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.7-flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.7-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-8-flash", "source_id": "google/gemini-3.8-flash", "name": "Gemini 3.8 Flash", "developer": "Google", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-8-flash", "description": "A generally available multimodal Gemini release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights; released September 2, 2026 as the current stable Flash model with no announced shutdown.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 3.8 Flash", "description": "Google's most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows", "family": "gemini-flash", "release_date": "2026-09-02", "last_updated": "2026-09-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides Gemini 3.8 Flash as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service. The terms permit customer applications but prohibit developing competing models and reverse engineering, extracting, or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to Gemini 3.8 Flash through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service; no weight license", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the stable release identified by the model code gemini-3.8-flash (generally available September 2, 2026; inputs text, image, video, audio, PDF; output text). No shutdown date is announced. The Cyber-only variant, the 3.7, 3.6, and 3.5 Flash siblings, the flash preview, the pro-preview line, image and audio-only releases, the Gemini consumer app, and the Vertex AI and AI Studio platforms themselves are separate model, service, or application boundaries.", "strengths": [ "Google documents a generally available model code with full modality matrix, thinking levels, token limits, release date, and knowledge cutoff", "Identity is corroborated across the API latest-model guide, the release changelog, the deprecations table, the models index, and a DeepMind model card" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release deprecates sampling parameters and thinking-budget configs, so existing integrations must migrate" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official guide: Gemini 3.8 Flash is generally available and ready for production", "url": "https://ai.google.dev/gemini-api/docs/latest-model", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official Enterprise Agent Platform Gemini 3.8 Flash page", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-8-flash", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official Gemini API deprecations table", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official release announcement (September 2, 2026)", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/google/gemini-3.8-flash.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12" }, { "id": "model-google-gemini-4-argon", "source_id": null, "name": "Gemini 4 Argon", "developer": "Google", "url": "https://deepmind.google/models/gemini/", "description": "Google's frontier Gemini release announced September 30, 2026, for long-horizon software engineering, enterprise knowledge work, and autonomous vulnerability remediation, with an expanded 1M-token output limit, no downloadable weights, and access staged from a restricted defender program to paid API customers.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini 4 Argon", "description": "Google's frontier model for real-world coding, enterprise knowledge work, and cyber defense", "family": "gemini-4", "release_date": "2026-09-30", "last_updated": "2026-09-30", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": null, "input": null, "output": 1000000 }, "reported_open_weights": false, "reported_license": null, "links": [ { "label": "Official release announcement", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/", "type": "announcement" } ], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google serves Gemini 4 Argon as a managed model under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. Those terms prohibit developing competing models and reverse engineering, extracting, or replicating model components such as parameter weights, so nothing is licensed for self-operation. At this review date no model identifier or public API path is published, so the terms are recorded as the governing framework for the announced paid API release rather than as a documented per-model license.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to Gemini 4 Argon through the Gemini API and Google AI Studio once released, including the incorporated Google APIs Terms of Service and Prohibited Use Policy; no weight license", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-30" } ], "access_boundary": "This record represents the Gemini 4 Argon release announced September 30, 2026, Google's frontier model for coding, enterprise knowledge work, and defensive cybersecurity, distinguished by a 1M-token output limit raised from 64K. At this review date Google has published no model identifier, no API model page, and no deprecation date: access is restricted to invited Fairwind Program defenders and is stated to reach developers, enterprises, and consumers starting with paid API customers. The Gemini 3.8 Flash and earlier Gemini releases, the Fairwind Program itself, the Gemini consumer app, and the Gemini Enterprise Agent Platform are separate service, program, or model boundaries.", "strengths": [ "First-party announcement and a DeepMind model page with a published benchmark table, safety posture, and eval methodology", "Announced introductory pricing of $2 per million input and $10 per million output tokens, with $4 and $20 applying after the introductory period" ], "tradeoffs": [ "No downloadable weights, and no published model identifier or API documentation at review time, so nothing can be pinned", "Access is staged: restricted to Fairwind Program defenders now, with no announced general-availability date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 2, "lifecycle_transparency": 4, "documentation_provenance": 7, "overall": 3.2 }, "evidence": [ { "kind": "web", "label": "Official Gemini 4 Argon announcement (September 30, 2026)", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Official DeepMind Gemini model page carrying the Gemini 4 Argon benchmark table, safety posture, and eval methodology", "url": "https://deepmind.google/models/gemini/", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Official Fairwind Program page describing restricted Gemini 4 Argon defender access", "url": "https://deepmind.google/fairwind-program/", "verified_at": "2026-09-30" }, { "kind": "web", "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-30" } ], "metadata_verified_at": "2026-09-30", "verified_at": "2026-09-30" }, { "id": "model-google-deep-research-preview-04-2026", "source_id": "google/deep-research-preview-04-2026", "name": "Gemini Deep Research Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/deep-research-preview-04-2026", "description": "An agentic research release served through Google's Interactions API under Google's API terms, with a 1,048,576-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini Deep Research Preview", "description": "Agentic model for autonomous multi-step research, synthesis, and cited reports", "family": "gemini-pro", "release_date": "2026-04-21", "last_updated": "2026-04-21", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": false }, "limits": { "context": 1048576, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Deep Research agent preview identified by the dated code deep-research-preview-04-2026, launched April 21, 2026 and still served. The December 2025 predecessor, the Max tier, the Interactions API platform, and later Deep Research releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with modalities, limits, and lifecycle status", "The deprecations table and changelog name the release and its migration path" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Serving runs on the paid Gemini API tiers through a single interface surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.26 }, "evidence": [ { "kind": "web", "label": "Official Deep Research agent model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/deep-research-preview-04-2026", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Deep Research launch announcement", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/next-generation-gemini-deep-research/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/deep-research-preview-04-2026.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-robotics-er-1-6-preview", "source_id": "google/gemini-robotics-er-1.6-preview", "name": "Gemini Robotics-ER 1.6 Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-robotics-er-1.6-preview", "description": "A retired embodied-reasoning preview release served through Google's Gemini API under Google's API terms, shut down August 31, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Gemini Robotics-ER 1.6 Preview", "description": "Vision-language model for embodied reasoning: spatial understanding, task planning, and physical-world agentic robotics", "family": "gemini", "release_date": "2026-04-14", "last_updated": "2026-04-14", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini Robotics-ER 1.6 preview identified by the model code gemini-robotics-er-1.6-preview, released April 14, 2026 and shut down August 31, 2026 with gemini-robotics-er-2-preview named as its replacement. The ER 2 line, the robotics action and model lines, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the release, deprecation, and shutdown dates in the changelog, deprecations table, and robotics overview", "The model page documents text output for plans, tool calls, and structured outputs" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is shut down and new traffic must migrate to the ER 2 preview" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini Robotics-ER 1.6 Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-robotics-er-1.6-preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API changelog documenting launch and shutdown", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-robotics-er-1.6-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemma-2-27b-it", "source_id": null, "name": "Gemma 2 27B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-2-27b-it", "description": "A gated downloadable 27B text-to-text instruction model governed by the Gemma Terms of Use and the incorporated Gemma Prohibited Use Policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Gemma 2 27B IT", "description": "Instruction-tuned 27B text-to-text decoder-only model from the Gemma 2 family, available in English", "family": "gemma", "release_date": "2024-06-27", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": null, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Gemma Terms of Use", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-2-27b-it" } ] }, "licenses": [ "LicenseRef-Gemma-Terms-of-Use", "LicenseRef-Gemma-Prohibited-Use-Policy" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The gated model card links the Gemma Terms of Use, whose appendix lists Gemma 2 and which incorporate the Gemma Prohibited Use Policy by reference. The terms require passing the use restrictions to every redistributor and reserve Google's right to restrict usage it believes violates them, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Gemma-Terms-of-Use", "scope": "Gemma 2 27B IT weights and tokenizer, with Gemma 2 named in the terms appendix", "kind": "web_terms", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-Gemma-Prohibited-Use-Policy", "scope": "Restricted-use policy incorporated by reference into the Gemma Terms of Use", "kind": "web_terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the google/gemma-2-27b-it release (initial Gemma 2 launch of June 27, 2024) at Hugging Face commit aaf20e6b9f4c0fcf043f6fb2a2068419086d77b0. The 27B pre-trained model, the 9B and 2B Gemma 2 sizes, Gemma 2 JPN, the Gemma 3 and Gemma 4 generations, quantizations, fine-tunes, and independent serving providers are separate boundaries.", "strengths": [ "Google publishes Transformers-compatible Safetensors shards with a tokenizer and names the governing terms on the model card", "Google's release log dates the Gemma 2 27B launch and the current terms list Gemma 2 in their appendix" ], "tradeoffs": [ "Artifact access is gated behind acceptance of Google's restricted Gemma Terms of Use", "The model card documents no hosted serving path beyond a link to Vertex Model Garden, and gives no context length, knowledge cutoff, or retirement note" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 6.1 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and Transformers guidance", "url": "https://huggingface.co/google/gemma-2-27b-it", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemma 2 model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_2", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemma Terms of Use governing the release", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemma Prohibited Use Policy incorporated into the terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemma release log dating the Gemma 2 27B launch", "url": "https://ai.google.dev/gemma/docs/releases", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-gemma-3-12b-it", "source_id": "google/gemma-3-12b-it", "name": "Gemma 3 12B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-3-12b-it", "description": "A gated 12B multimodal instruction model handling text and image input with a 128K-token context, governed by the Gemma Terms of Use.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Gemma 3 12B IT", "description": "Open multimodal Gemma instruction model for multilingual text generation and image understanding", "family": "gemma", "release_date": "2025-03-12", "last_updated": "2025-03-12", "knowledge_cutoff": "2024-08", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3-12b-it" } ] }, "licenses": [ "LicenseRef-Gemma-Terms-of-Use", "LicenseRef-Gemma-Prohibited-Use-Policy" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repositories ship no LICENSE file; the gated model cards link the governing Gemma Terms of Use, whose appendix lists Gemma 3 and which incorporate the Gemma Prohibited Use Policy by reference. The terms restrict use rather than grant open source rights, so these releases are source-available, in contrast to the Gemma 4 generation's Apache-2.0 carve-out.", "license_evidence": [ { "license_id": "LicenseRef-Gemma-Terms-of-Use", "scope": "Official google/gemma-3-12b-it repository, Safetensors weights, tokenizer, processor, and chat template", "kind": "web_terms", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Gemma-Prohibited-Use-Policy", "scope": "Restricted-use policy incorporated by reference into the Gemma Terms of Use", "kind": "web_terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the google/gemma-3-12b-it release of March 2025 at Hugging Face commit 96b6f1eccf38110c56df3a15bffe176da04bfd80. The other Gemma 3 sizes, the Gemma 4 generation, quantizations, fine-tunes, and independent serving providers remain separate boundaries.", "strengths": [ "Dated March 2025 release with pinned repositories and a full technical report", "Google documents Transformers, vLLM, SGLang, Ollama, llama.cpp, MLX, and on-device runtimes" ], "tradeoffs": [ "Artifact access is gated behind acceptance of Google's restricted Gemma Terms of Use", "Google no longer documents served copies of Gemma 3 on the Gemini API and treats the line as legacy without a retirement note" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 6.24 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/google/gemma-3-12b-it", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma Terms of Use governing the release", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma Prohibited Use Policy incorporated into the terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemma-3-12b-it.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemma-3-27b-it", "source_id": "google/gemma-3-27b-it", "name": "Gemma 3 27B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-3-27b-it", "description": "A gated 27B multimodal instruction model handling text and image input with a 128K-token context, governed by the Gemma Terms of Use.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Gemma 3 27B IT", "description": "Largest open Gemma 3 instruction model for multilingual text generation and visual understanding", "family": "gemma", "release_date": "2025-03-12", "last_updated": "2025-03-12", "knowledge_cutoff": "2024-08", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3-27b-it" } ] }, "licenses": [ "LicenseRef-Gemma-Terms-of-Use", "LicenseRef-Gemma-Prohibited-Use-Policy" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repositories ship no LICENSE file; the gated model cards link the governing Gemma Terms of Use, whose appendix lists Gemma 3 and which incorporate the Gemma Prohibited Use Policy by reference. The terms restrict use rather than grant open source rights, so these releases are source-available, in contrast to the Gemma 4 generation's Apache-2.0 carve-out.", "license_evidence": [ { "license_id": "LicenseRef-Gemma-Terms-of-Use", "scope": "Official google/gemma-3-27b-it repository, Safetensors weights, tokenizer, processor, and chat template", "kind": "web_terms", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Gemma-Prohibited-Use-Policy", "scope": "Restricted-use policy incorporated by reference into the Gemma Terms of Use", "kind": "web_terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the google/gemma-3-27b-it release of March 2025 at Hugging Face commit 005ad3404e59d6023443cb575daa05336842228a. The other Gemma 3 sizes, the Gemma 4 generation, quantizations, fine-tunes, and independent serving providers remain separate boundaries.", "strengths": [ "Dated March 2025 release with pinned repositories and a full technical report", "Google documents Transformers, vLLM, SGLang, Ollama, llama.cpp, MLX, and on-device runtimes" ], "tradeoffs": [ "Artifact access is gated behind acceptance of Google's restricted Gemma Terms of Use", "Google no longer documents served copies of Gemma 3 on the Gemini API and treats the line as legacy without a retirement note" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 6.24 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/google/gemma-3-27b-it", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma Terms of Use governing the release", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma Prohibited Use Policy incorporated into the terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemma-3-27b-it.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemma-3-4b-it", "source_id": "google/gemma-3-4b-it", "name": "Gemma 3 4B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-3-4b-it", "description": "A gated 4B multimodal instruction model handling text and image input with a 128K-token context, governed by the Gemma Terms of Use.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Gemma 3 4B IT", "description": "Open multimodal Gemma instruction model for efficient text generation and image understanding", "family": "gemma", "release_date": "2025-03-12", "last_updated": "2025-03-12", "knowledge_cutoff": "2024-08", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3-4b-it" } ] }, "licenses": [ "LicenseRef-Gemma-Terms-of-Use", "LicenseRef-Gemma-Prohibited-Use-Policy" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repositories ship no LICENSE file; the gated model cards link the governing Gemma Terms of Use, whose appendix lists Gemma 3 and which incorporate the Gemma Prohibited Use Policy by reference. The terms restrict use rather than grant open source rights, so these releases are source-available, in contrast to the Gemma 4 generation's Apache-2.0 carve-out.", "license_evidence": [ { "license_id": "LicenseRef-Gemma-Terms-of-Use", "scope": "Official google/gemma-3-4b-it repository, Safetensors weights, tokenizer, processor, and chat template", "kind": "web_terms", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Gemma-Prohibited-Use-Policy", "scope": "Restricted-use policy incorporated by reference into the Gemma Terms of Use", "kind": "web_terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the google/gemma-3-4b-it release of March 2025 at Hugging Face commit 093f9f388b31de276ce2de164bdc2081324b9767. The other Gemma 3 sizes, the Gemma 4 generation, quantizations, fine-tunes, and independent serving providers remain separate boundaries.", "strengths": [ "Dated March 2025 release with pinned repositories and a full technical report", "Google documents Transformers, vLLM, SGLang, Ollama, llama.cpp, MLX, and on-device runtimes" ], "tradeoffs": [ "Artifact access is gated behind acceptance of Google's restricted Gemma Terms of Use", "Google no longer documents served copies of Gemma 3 on the Gemini API and treats the line as legacy without a retirement note" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 6.24 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/google/gemma-3-4b-it", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma Terms of Use governing the release", "url": "https://ai.google.dev/gemma/terms", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma Prohibited Use Policy incorporated into the terms", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemma-3-4b-it.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemma-4-12b-it", "source_id": "google/gemma-4-12b-it", "name": "Gemma 4 12B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-4-12B-it", "description": "Google's Gemma 4 12B instruction-tuned unified multimodal release of June 3, 2026, an encoder-free model accepting text, image, audio, and video input with 256K context, published as Apache-2.0 open weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Gemma 4 12B IT", "description": "Compact Gemma 4 instruction model for open, self-hosted chat and reasoning", "family": "gemma", "release_date": "2026-06-09", "last_updated": "2026-06-09", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-12B-it" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card and Google's Gemma 4 license page state Apache-2.0 as the governing license; the repository ships no separate LICENSE file, matching the reviewed Gemma 4 31B IT record's evidence pattern. models.dev's release date and text-only modalities understate the first-party release.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official google/gemma-4-12B-it model repository, Safetensors weights, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/google/gemma-4-12B-it", "verified_at": "2026-09-22" }, { "license_id": "Apache-2.0", "scope": "Google's Gemma 4 license page, linked from the model card as the governing license", "kind": "web_terms", "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the google/gemma-4-12B-it instruction-tuned release of June 3, 2026. The Gemma 4 E2B, E4B, 26B A4B, and 31B variants, the base checkpoint, quantizations, and hosted Gemma services are separate boundaries.", "strengths": [ "Ungated Safetensors weights on the official google org with an encoder-free unified multimodal architecture", "Clear Apache-2.0 grant stated by the model card and Google's Gemma 4 license page" ], "tradeoffs": [ "The repository ships no LICENSE file; the grant rests on model-card metadata and the license page", "models.dev's discovery metadata understates the release date and the audio and video input modalities" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.06 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/google/gemma-4-12b-it.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/google/gemma-4-12B-it", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Gemma 4 12B announcement and technical report", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-google-gemma-4-26b-a4b-it", "source_id": "google/gemma-4-26b-a4b-it", "name": "Gemma 4 26B A4B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it", "description": "An ungated 25.8B mixture-of-experts multimodal instruction model handling text and image input with a 256K-token context, released under Apache-2.0 and served on the Gemini API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Gemma 4 26B A4B IT", "description": "Open Gemma instruction model for efficient chat and self-hosted deployments", "family": "gemma", "release_date": "2026-04-02", "last_updated": "2026-04-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card and Google's Gemma 4 license page apply Apache-2.0 to the published weights, tokenizer, and chat template. The Gemma Terms of Use that govern earlier Gemma generations explicitly refer Gemma 4 to the separate Apache-2.0 license, so no use-restriction addendum applies to this release.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "verified_at": "2026-09-18", "scope": "Official google/gemma-4-26B-A4B-it model repository, Safetensors weights, tokenizer, processor, and chat template", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it/blob/01e5b3ee840d3a9e0b0b493c593e85398a30ef75/README.md" }, { "license_id": "Apache-2.0", "kind": "web_terms", "verified_at": "2026-09-18", "scope": "Google's Gemma 4 license page, linked from the model card as the governing license text", "url": "https://ai.google.dev/gemma/apache_2" } ], "access_boundary": "This record represents the google/gemma-4-26B-A4B-it release of April 2026 at Hugging Face commit head. The other Gemma 4 sizes, the Gemma 3 generation, quantizations, fine-tunes, and independent serving providers remain separate boundaries.", "strengths": [ "Ungated Apache-2.0 weights with the license page linked from the model card", "Dated April 2026 release with a model card, Google model-card page, and technical report" ], "tradeoffs": [ "No LICENSE file ships in the repository; the grant rests on the card metadata and license page", "No deprecation or update posture is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.2 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma 4 Apache-2.0 license page", "url": "https://ai.google.dev/gemma/apache_2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemma-4-26b-a4b-it.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemma-4-e2b-it", "source_id": "google/gemma-4-E2B-it", "name": "Gemma 4 E2B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-4-E2B-it", "description": "An ungated on-device 5.1B multimodal instruction model handling text, image, and audio input with a 128K-token context, released under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Gemma 4 E2B IT", "description": "Open Gemma instruction model for efficient chat and self-hosted deployments", "family": "gemma", "release_date": "2026-04-02", "last_updated": "2026-04-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-E2B-it" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card and Google's Gemma 4 license page apply Apache-2.0 to the published weights, tokenizer, and chat template. The Gemma Terms of Use that govern earlier Gemma generations explicitly refer Gemma 4 to the separate Apache-2.0 license, so no use-restriction addendum applies to this release.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "verified_at": "2026-09-18", "scope": "Official google/gemma-4-E2B-it model repository, Safetensors weights, tokenizer, processor, and chat template", "url": "https://huggingface.co/google/gemma-4-E2B-it/blob/179516f0c449474fdc46f08f30ead5b11e178497/README.md" }, { "license_id": "Apache-2.0", "kind": "web_terms", "verified_at": "2026-09-18", "scope": "Google's Gemma 4 license page, linked from the model card as the governing license text", "url": "https://ai.google.dev/gemma/apache_2" } ], "access_boundary": "This record represents the google/gemma-4-E2B-it release of April 2026 at Hugging Face commit head. The other Gemma 4 sizes, the Gemma 3 generation, quantizations, fine-tunes, and independent serving providers remain separate boundaries.", "strengths": [ "Ungated Apache-2.0 weights with the license page linked from the model card", "Dated April 2026 release with a model card, Google model-card page, and technical report" ], "tradeoffs": [ "No LICENSE file ships in the repository; the grant rests on the card metadata and license page", "No deprecation or update posture is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 4, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.84 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/google/gemma-4-E2B-it", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma 4 Apache-2.0 license page", "url": "https://ai.google.dev/gemma/apache_2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemma-4-E2B-it.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemma-4-e4b-it", "source_id": "google/gemma-4-E4B-it", "name": "Gemma 4 E4B IT", "developer": "Google", "url": "https://huggingface.co/google/gemma-4-E4B-it", "description": "An ungated on-device 8B multimodal instruction model handling text, image, and audio input with a 128K-token context, released under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Gemma 4 E4B IT", "description": "Open Gemma instruction model for efficient chat and self-hosted deployments", "family": "gemma", "release_date": "2026-04-02", "last_updated": "2026-04-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-E4B-it" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card and Google's Gemma 4 license page apply Apache-2.0 to the published weights, tokenizer, and chat template. The Gemma Terms of Use that govern earlier Gemma generations explicitly refer Gemma 4 to the separate Apache-2.0 license, so no use-restriction addendum applies to this release.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "verified_at": "2026-09-18", "scope": "Official google/gemma-4-E4B-it model repository, Safetensors weights, tokenizer, processor, and chat template", "url": "https://huggingface.co/google/gemma-4-E4B-it/blob/fa62d88df2e6df5efa9d26ad6b3beaea2765f0cd/README.md" }, { "license_id": "Apache-2.0", "kind": "web_terms", "verified_at": "2026-09-18", "scope": "Google's Gemma 4 license page, linked from the model card as the governing license text", "url": "https://ai.google.dev/gemma/apache_2" } ], "access_boundary": "This record represents the google/gemma-4-E4B-it release of April 2026 at Hugging Face commit head. The other Gemma 4 sizes, the Gemma 3 generation, quantizations, fine-tunes, and independent serving providers remain separate boundaries.", "strengths": [ "Ungated Apache-2.0 weights with the license page linked from the model card", "Dated April 2026 release with a model card, Google model-card page, and technical report" ], "tradeoffs": [ "No LICENSE file ships in the repository; the grant rests on the card metadata and license page", "No deprecation or update posture is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 4, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.84 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/google/gemma-4-E4B-it", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemma 4 Apache-2.0 license page", "url": "https://ai.google.dev/gemma/apache_2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemma-4-E4B-it.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-lyria-3-clip-preview", "source_id": "google/lyria-3-clip-preview", "name": "Lyria 3 Clip Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/lyria-3-clip-preview", "description": "Google's Lyria 3 Clip preview, a music generation model for 30-second clips and loops from text or image prompts, served through the Gemini API under Google's API terms as MP3 audio with lyrics text and no downloadable model weights.", "model_type": "audio_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Lyria 3 Clip Preview", "description": "Music generation model for short 30-second clips, loops, and previews from text or image prompts", "family": "lyria", "release_date": "2026-03-25", "last_updated": "2026-03-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text", "audio" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": false, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Lyria 3 Clip preview identified by the Gemini API model code lyria-3-clip-preview, launched March 25, 2026 with no shutdown date announced. Its documented output is audio (MP3) with lyrics as text; the record is classed by the audio it exists to produce. Lyria 3 Pro Preview, the generally available Lyria 3.5 of September 3, 2026, Lyria RealTime, and the Vertex-based Lyria products remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with the model code, inputs, outputs, and a version line marking it as a preview", "The changelog and deprecations table name the launch date and state that no shutdown date is announced" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The deprecations table lists no shutdown date for this preview and the model index now leads with the generally available Lyria 3.5 line, so its continued availability is open" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 5, "documentation_provenance": 8, "overall": 4.16 }, "evidence": [ { "kind": "web", "label": "Official Lyria 3 Clip Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/lyria-3-clip-preview", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API changelog documenting the Lyria 3 launch", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API model index listing Lyria 3 Clip beside Lyria 3.5", "url": "https://ai.google.dev/gemini-api/docs/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/google/lyria-3-clip-preview.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-lyria-3-pro-preview", "source_id": "google/lyria-3-pro-preview", "name": "Lyria 3 Pro Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/lyria-3-pro-preview", "description": "Google's Lyria 3 Pro preview, a music generation model for full-length songs with verses, choruses, and bridges from text or image prompts, served through the Gemini API under Google's API terms as MP3 audio with lyrics text and no downloadable model weights.", "model_type": "audio_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Lyria 3 Pro Preview", "description": "Music generation model for full-length songs from text or images with vocals and structure", "family": "lyria", "release_date": "2026-03-25", "last_updated": "2026-03-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text", "audio" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": false, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Lyria 3 Pro preview identified by the Gemini API model code lyria-3-pro-preview, launched March 25, 2026 with no shutdown date announced and lyria-3.5 listed as the recommended replacement. Its documented output is audio (MP3) with lyrics as text; the record is classed by the audio it exists to produce. Lyria 3 Clip Preview, the generally available Lyria 3.5, Lyria RealTime, and the Vertex-based Lyria products remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page, a changelog launch entry, and a deprecations row with a named replacement", "The model index labels it the previous-generation song model, so its position in the line is stated" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The model page's data table lists lyria-3-clip-preview as its model code, an inconsistency in Google's own page, and no shutdown date is announced for the superseded preview" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 4.02 }, "evidence": [ { "kind": "web", "label": "Official Lyria 3 Pro Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/lyria-3-pro-preview", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API changelog documenting the Lyria 3 launch", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API model index describing Lyria 3 Pro as the previous generation", "url": "https://ai.google.dev/gemini-api/docs/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/google/lyria-3-pro-preview.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-gemini-3-1-flash-image", "source_id": "google/gemini-3.1-flash-image", "name": "Nano Banana 2", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image", "description": "An active image generation and editing release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nano Banana 2", "description": "Image model for prompt-driven generation, editing, and visual design workflows", "family": "gemini-flash", "release_date": "2026-05-28", "last_updated": "2026-05-28", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3.1 Flash Image (Nano Banana 2) generally available release of May 2026 under the model code gemini-3.1-flash-image. The shut-down preview endpoint, the Pro and Lite image lines, the Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the model page, model card, GA changelog entry, and a dated retirement horizon", "The launch announcement names the Gemini Enterprise Agent Platform availability" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Serving runs on the paid Gemini API tiers through a single interface surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.1 Flash Image model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API changelog documenting the GA release", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GA announcement", "url": "https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.1-flash-image.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-1-flash-lite-image", "source_id": "google/gemini-3.1-flash-lite-image", "name": "Nano Banana 2 Lite", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-image", "description": "An active image generation and editing release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nano Banana 2 Lite", "description": "Fastest, most cost-efficient Gemini image model for high-volume 1K generation and editing", "family": "gemini-flash-lite", "release_date": "2026-06-30", "last_updated": "2026-06-30", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 4096 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3.1 Flash-Lite Image (Nano Banana 2 Lite) release of June 2026 under the model code gemini-3.1-flash-lite-image, launched straight to general availability. The Pro and Flash image lines, the Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the model page, model card, and GA changelog entry with a dated retirement horizon", "The launch announcement names it the recommended replacement for Nano Banana (gemini-2.5-flash-image)" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Serving runs on the paid Gemini API tiers through a single interface surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3.1 Flash-Lite Image model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-image", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API changelog documenting the GA release", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official launch announcement", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-flash-nano-banana-2-lite/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.1-flash-lite-image.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-1-flash-image-preview", "source_id": "google/gemini-3.1-flash-image-preview", "name": "Nano Banana 2 Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "description": "A retired image generation and editing preview release served through Google's Gemini API under Google's API terms, shut down June 25, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nano Banana 2 Preview", "description": "Image model for prompt-driven generation, editing, and visual design workflows", "family": "gemini-flash", "release_date": "2026-02-26", "last_updated": "2026-02-26", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3.1 Flash Image Preview (Nano Banana 2 Preview) release of February 2026, shut down June 25, 2026 with the generally available gemini-3.1-flash-image as its replacement. The GA line, the Pro and Lite image lines, the Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the release, deprecation, and shutdown dates in the changelog and deprecations table", "The Agent Platform release notes confirm the retirement and name the migration targets" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is shut down; its documentation page is no longer served" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API changelog documenting launch and shutdown", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Nano Banana 2 launch announcement", "url": "https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3.1-flash-image-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-pro-image", "source_id": "google/gemini-3-pro-image", "name": "Nano Banana Pro", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image", "description": "An active image generation and editing release served through Google's Gemini API and Vertex AI under Google's API terms, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nano Banana Pro", "description": "Nano Banana Pro for higher-fidelity image generation and design-heavy edits", "family": "gemini-pro", "release_date": "2026-05-28", "last_updated": "2026-05-28", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3 Pro Image (Nano Banana Pro) generally available release of May 2026 under the model code gemini-3-pro-image. The shut-down preview endpoint, the Flash and Lite image lines, the Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the model page, model card, GA changelog entry, and a dated retirement horizon", "The launch announcement names the Gemini Enterprise Agent Platform availability" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Serving runs on the paid Gemini API tiers through a single interface surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official Gemini 3 Pro Image model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API changelog documenting the GA release", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GA announcement", "url": "https://blog.google/innovation-and-ai/products/nano-banana-pro/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3-pro-image.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-gemini-3-pro-image-preview", "source_id": "google/gemini-3-pro-image-preview", "name": "Nano Banana Pro Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "description": "A retired image generation and editing preview release served through Google's Gemini API under Google's API terms, shut down June 25, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Nano Banana Pro Preview", "description": "Nano Banana Pro for higher-fidelity image generation and design-heavy edits", "family": "gemini-pro", "release_date": "2025-11-20", "last_updated": "2025-11-20", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image" ], "output": [ "text", "image" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-09-18", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy" } ], "access_boundary": "This record represents the Gemini 3 Pro Image Preview (Nano Banana Pro Preview) release of November 2025, shut down June 25, 2026 with the generally available gemini-3-pro-image as its replacement. The GA line, the Flash and Lite image lines, the Gemini app products, and later Gemini releases remain separate model or service boundaries.", "strengths": [ "Google documents the release, deprecation, and shutdown dates in the changelog and deprecations table", "The launch announcement names the model and its paid preview rollout" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is shut down; its documentation page is no longer served" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Gemini API changelog documenting launch and shutdown", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Nano Banana Pro launch announcement", "url": "https://blog.google/innovation-and-ai/products/nano-banana-pro/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/google/gemini-3-pro-image-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-google-veo-3-1-lite-generate-preview", "source_id": "google/veo-3.1-lite-generate-preview", "name": "Veo 3.1 Lite Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/veo-3.1-lite-generate-preview", "description": "Google's Veo 3.1 Lite video generation preview release, the lowest-cost Veo 3.1 tier, producing video with synchronized audio from text or image prompts through the Gemini API under Google's API terms, with no downloadable model weights and a documented earliest shutdown of October 22, 2026.", "model_type": "video_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Veo 3.1 Lite Preview", "description": "Video model for prompt-guided generation, editing, and motion workflows", "family": "veo", "release_date": "2026-03-31", "last_updated": "2026-03-31", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "video" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 1024, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Veo 3.1 Lite preview release identified by the Gemini API model code veo-3.1-lite-generate-preview, launched March 31, 2026 and listed with an earliest shutdown date of October 22, 2026 and gemini-omni-1.1-flash as its recommended replacement. Google documents that it supports neither 4K output nor extension. The standard and Fast Veo 3.1 previews, the Gemini Omni Flash line, and later video releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with the model code, supported inputs, output type, and a latest-update month", "The changelog and deprecations table give the launch date, an earliest shutdown date, and a named replacement" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is a preview with a documented earliest shutdown of October 22, 2026, so continued availability depends on Google's lifecycle decisions" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Veo 3.1 Lite Preview model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/veo-3.1-lite-generate-preview", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API changelog documenting the Veo 3.1 Lite Preview launch", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/google/veo-3.1-lite-generate-preview.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-veo-3-1-generate-preview", "source_id": "google/veo-3.1-generate-preview", "name": "Veo 3.1 Preview", "developer": "Google", "url": "https://ai.google.dev/gemini-api/docs/models/veo-3.1-generate-preview", "description": "Google's Veo 3.1 video generation preview release, producing video with synchronized audio from text or image prompts through the Gemini API under Google's API terms, with no downloadable model weights and a documented earliest shutdown of October 22, 2026.", "model_type": "video_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Veo 3.1 Preview", "description": "Video model for prompt-guided generation, editing, and motion workflows", "family": "veo", "release_date": "2025-10-15", "last_updated": "2026-01", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "video" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 1024, "input": null, "output": 1 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Gemini-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Google provides the release as a managed service under the Gemini API Additional Terms of Service, which incorporate the Google APIs Terms of Service and a Prohibited Use Policy. The terms permit customer applications but prohibit extracting or replicating model components such as parameter weights, so nothing is licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Gemini-API-Terms", "scope": "Access to the release through the Gemini API and Google AI Studio, including the incorporated Google APIs Terms of Service and Prohibited Use Policy", "kind": "web_terms", "url": "https://ai.google.dev/gemini-api/terms", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Veo 3.1 preview release identified by the Gemini API model code veo-3.1-generate-preview, released October 15, 2025 and listed with an earliest shutdown date of October 22, 2026 and gemini-omni-1.1-flash as its recommended replacement. The Veo 3.1 Fast and Veo 3.1 Lite previews, the generally available veo-3.1-generate-001 on the Gemini Enterprise Agent Platform, the Gemini Omni Flash line, and later video releases remain separate model or service boundaries.", "strengths": [ "Google documents a release-specific model page with the model code, supported inputs, video-with-audio output, and a latest-update month", "The deprecations table lists the release date, an earliest shutdown date, and a named replacement" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is a preview with a documented earliest shutdown of October 22, 2026, so continued availability depends on Google's lifecycle decisions" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official Veo 3.1 model documentation naming veo-3.1-generate-preview", "url": "https://ai.google.dev/gemini-api/docs/models/veo-3.1-generate-preview", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API deprecations and model lifecycle page", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Gemini API changelog documenting the Veo 3.1 preview launch", "url": "https://ai.google.dev/gemini-api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/google/veo-3.1-generate-preview.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-google-gemma-4-31b-it", "source_id": "google/gemma-4-31b-it", "name": "Gemma 4 31B IT", "developer": "Google DeepMind", "url": "https://huggingface.co/google/gemma-4-31B-it", "description": "A downloadable image-and-text instruction-tuned Gemma 4 release published by Google DeepMind under Apache-2.0, with Safetensors weights that are not gated.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Gemma 4 31B IT", "description": "Largest Gemma 4 instruction model for open, self-hosted chat and reasoning", "family": "gemma", "release_date": "2026-04-02", "last_updated": "2026-04-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-31B-it" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card and Google's Gemma 4 license page apply Apache-2.0 to the published gemma-4-31B-it weights, tokenizer, and chat template. The Gemma Terms of Use and Gemma Prohibited Use Policy that govern earlier Gemma generations explicitly refer Gemma 4 to the separate Apache-2.0 license, so no use-restriction addendum applies to this release.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official google/gemma-4-31B-it model repository, Safetensors weights, tokenizer, processor, and chat template", "kind": "web_terms", "url": "https://huggingface.co/google/gemma-4-31B-it/blob/842da3794eaa0b77d5f08bae87a17459d91ff475/README.md", "verified_at": "2026-09-11" }, { "license_id": "Apache-2.0", "scope": "Google's Gemma 4 license page, linked from the model card as the governing license text", "kind": "web_terms", "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the google/gemma-4-31B-it instruction-tuned release. The gemma-4-31B pre-trained base model, the E2B, E4B, 12B, and 26B A4B variants, the -it-assistant repositories, quantized or QAT checkpoints, the Gemini API and Vertex AI served copies, and later Gemma releases remain separate model or service boundaries.", "strengths": [ "Google publishes ungated Safetensors shards, an index, tokenizer, processor, and chat template files under Apache-2.0, plus a technical report", "The official model card documents Transformers, vLLM, SGLang, and Docker Model Runner paths, and Google documents Gemini API, Vertex AI Model Garden, GKE, Cloud Run, Ollama, llama.cpp, and MLX paths" ], "tradeoffs": [ "The license is stated in model-card metadata and on a Google page rather than in a repository LICENSE file, and Google's Gemma site still links the older Terms of Use and Prohibited Use Policy alongside it", "No deprecation or update posture is documented for the downloadable release, and the served Gemini API copy follows separate API terms" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.06 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/google/gemma-4-31B-it", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Gemma 4 documentation with download and deployment paths", "url": "https://ai.google.dev/gemma/docs/core", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/google/gemma-4-31b-it.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-ibm-granite-4-h-micro", "source_id": "ibm/granite-4-h-micro", "name": "Granite-4.0-H-Micro", "developer": "IBM", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro", "description": "IBM's Granite-4.0-H-Micro, a 3-billion-parameter hybrid text instruction model with 128K context and tool-calling, published with signed Safetensors checkpoints under Apache-2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Granite-4.0-H-Micro", "description": "Compact open-weight hybrid Granite model for lightweight enterprise chat and tool calling", "family": "granite", "release_date": "2025-10-02", "last_updated": "2025-10-02", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official IBM model card and the IBM Granite release materials identify Apache 2.0 as the license for the published Granite-4.0-H-Micro repository and checkpoint, mirroring the reviewed Granite-4.0-H-Small record. The repository carries the license identifier in its pinned model card rather than a separate license file.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official granite-4.0-h-micro model repository, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the ibm-granite/granite-4.0-h-micro instruct release dated October 2, 2025. The Granite-4.0-H-Micro-Base model, the Tiny, Small, and 8B variants, quantizations, IBM watsonx.ai services, and independent hosts are separate boundaries.", "strengths": [ "IBM publishes ungated signed Safetensors shards under an OSI-approved Apache-2.0 grant", "Serving is documented through watsonx.ai and several independent distribution and hosting paths" ], "tradeoffs": [ "The license statement is carried in model-card metadata instead of a repository license file", "The hybrid GQA-Mamba2 architecture is documented only through the Transformers path, so other runtimes need independent verification" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.86 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/ibm/granite-4-h-micro.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-ibm-granite-4-h-small", "source_id": "ibm/granite-4-h-small", "name": "Granite-4.0-H-Small", "developer": "IBM", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-small", "description": "A downloadable hybrid text instruction model published by IBM with signed Safetensors checkpoints and an Apache-2.0 distribution boundary.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Granite-4.0-H-Small", "description": "Open-weight hybrid model for enterprise chat, coding, retrieval-augmented generation, and tool-calling workloads", "family": "granite", "release_date": "2025-10-02", "last_updated": "2025-10-02", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-small" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official IBM model card identifies Apache 2.0 as the license for the published Granite-4.0-H-Small repository and checkpoint, and IBM's release announcement confirms the Apache 2.0 grant for all Granite 4.0 models. The repository carries the license identifier in its pinned model card rather than a separate license file.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official granite-4.0-h-small model repository, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-small/blob/b8c0982bab7fde4eb48110f5a069527c008fab39/README.md", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the ibm-granite/granite-4.0-h-small instruct release dated October 2, 2025. The Granite-4.0-H-Small-Base model, the Tiny, Micro, and 8B variants, quantizations, IBM watsonx.ai services, and independent hosts are separate model or service boundaries.", "strengths": [ "IBM publishes ungated Safetensors shards with a model.sig signature file under an OSI-approved license", "IBM's release announcement names watsonx.ai and several independent distribution and hosting paths for the release" ], "tradeoffs": [ "The license statement is carried in model-card metadata instead of a repository license file", "The model card documents only a Transformers path for the hybrid architecture, so other runtimes need independent verification" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.86 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and Transformers guidance", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-small", "verified_at": "2026-09-11" }, { "kind": "web", "label": "IBM Granite 4.0 release announcement with license, signing, and availability statements", "url": "https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/ibm/granite-4-h-small.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-ibm-granite-4-2-8b", "source_id": null, "name": "Granite-4.2-8B", "developer": "IBM", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b", "description": "A downloadable 8B dense reasoning text model with tool calling published by IBM under the Apache-2.0 license, with signed Safetensors checkpoints.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Granite-4.2-8B", "description": "8B decoder-only dense reasoning model with switchable thinking modes and tool calling, built on Granite-4.1-8B-Base, natively 128K context extendable to 512K", "family": "granite", "release_date": "2026-08-25", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 131072, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official IBM model card and its metadata name Apache 2.0 as the license for the published Granite-4.2-8B repository. The repository carries the license in its pinned model card rather than a separate license file, as with other Granite releases.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official granite-4.2-8b model repository, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b/blob/f8de16cdcdbc6c779ca517604e050d82cc119e44/README.md", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the ibm-granite/granite-4.2-8b instruct reasoning release dated August 25, 2026 at Hugging Face commit f8de16cdcdbc6c779ca517604e050d82cc119e44. The Granite-4.1-8B-Base model it derives from, the 3B and 30B Granite 4.2 siblings, the Granite 4.0 generation, quantizations, IBM-hosted services, and independent hosts are separate boundaries.", "strengths": [ "IBM publishes ungated Safetensors shards with a model.sig signature file under an OSI-approved license", "The model card documents vLLM and SGLang serving, a custom thinking parser, and a dated release" ], "tradeoffs": [ "The license statement is carried in model-card metadata and a badge rather than a repository license file", "The model card documents no IBM-hosted serving path, so managed availability must be verified separately" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.38 }, "evidence": [ { "kind": "web", "label": "Official model card, license statement, and serving guidance", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Granite 4.2 technical blog", "url": "https://huggingface.co/blog/ibm-granite/granite-4-2", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official repository metadata listing the signature file and Safetensors shards", "url": "https://huggingface.co/api/models/ibm-granite/granite-4.2-8b", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-inclusionai-ling-3-0-flash", "source_id": null, "name": "Ling-3.0-flash", "developer": "Inclusion AI (Ant Group)", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash", "description": "Inclusion AI's Ling-3.0-flash, the base release of the Ling 3.0 flash line: a native hybrid-reasoning mixture-of-experts text model with 124B total and 5.1B active parameters and a 256K context, published as open weights under MIT.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ling-3.0-flash", "description": "Native hybrid reasoning model with 124B total and 5.1B active parameters and hybrid linear attention, aimed at agentic coding, deep research, and long-context work", "family": "ling", "release_date": null, "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MIT", "links": [ { "label": "Hugging Face model card", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card metadata declares MIT. The repository ships no separate LICENSE file (the raw LICENSE path returns 404 on 2026-10-06), so the grant rests on the card metadata alone, as for the reviewed Ling-3.0-flash-Fin release.", "license_evidence": [ { "license_id": "MIT", "scope": "Ling-3.0-flash repository weights, Safetensors checkpoints, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Ling-3.0-flash BF16 weight release in the inclusionAI/Ling-3.0-flash repository, the base of the flash line. The Fin and VL continuations, the fp8, fp4, int4 and GGUF siblings, base and midtrain checkpoints, and any gateway-hosted free variant are separate boundaries.", "strengths": [ "Ungated BF16 weights with SGLang and vLLM launch recipes and a linked training-content summary", "Documented 124B total and 5.1B active parameter architecture with a 256K context schedule" ], "tradeoffs": [ "The MIT grant rests on model-card metadata with no LICENSE file in the repository", "No first-party hosted API for the release was found in the sources read, and the card documents no release date" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.6 }, "evidence": [ { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official inclusionAI Hugging Face organization page", "url": "https://huggingface.co/inclusionAI", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-inclusionai-ling-3-0-flash-fin", "source_id": "inclusionai/ling-3.0-flash-fin", "name": "Ling-3.0-flash-Fin", "developer": "Inclusion AI (Ant Group)", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin", "description": "Inclusion AI's Ling-3.0-flash-Fin, the finance-specialized continuation of the Ling 3.0 flash line with 124B total and 5.1B active parameters, 256K context, and the FinFIRST research evaluation, published as open weights under MIT.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Ling 3.0 Flash Fin", "description": "Finance-enhanced model for financial research, multi-step investment workflows, and long-horizon planning and execution", "family": "ling", "release_date": "2026-08-27", "last_updated": "2026-08-27", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card declares MIT and the base Ling-3.0-flash card declares the same; the repository ships no separate LICENSE file, so the grant rests on the pinned model-card metadata. models.dev's row reports open weights as false, which is incorrect for this ungated release.", "license_evidence": [ { "license_id": "MIT", "scope": "Ling-3.0-flash-Fin repository weights, Safetensors checkpoints, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Ling-3.0-flash-Fin weight release, a continued-train from the Ling-3.0-flash base. The base release, fp8 and int4 quantized siblings, and OpenRouter-hosted copies are separate boundaries.", "strengths": [ "First finance-specialized Ling release with the FinFIRST research evaluation and source-grounded financial focus", "Ungated BF16 weights with explicit SGLang and vLLM serving guidance" ], "tradeoffs": [ "The MIT grant rests on model-card metadata with no LICENSE file in the repository", "A continued-train of the flash base, so its boundary excludes the base, quantized siblings, and API copies" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.0 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/inclusionai/ling-3.0-flash-fin.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-inclusionai-ling-3-0-flash-vl", "source_id": null, "name": "Ling-3.0-flash-VL", "developer": "Inclusion AI (Ant Group)", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL", "description": "Inclusion AI's Ling-3.0-flash-VL, the native multimodal continuation of Ling-3.0-flash that reads image and video input and writes text, with 124B total and 5.5B active parameters and a 256K context, published as open weights under MIT.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ling-3.0-flash-VL", "description": "Native multimodal model built on Ling-3.0-flash with image and video understanding, reasoning, and interface-acting capability", "family": "ling", "release_date": null, "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": null, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MIT", "links": [ { "label": "Hugging Face model card", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card metadata declares MIT. The repository ships no separate LICENSE file (the raw LICENSE path returns 404 on 2026-10-06), so the grant rests on the card metadata alone.", "license_evidence": [ { "license_id": "MIT", "scope": "Ling-3.0-flash-VL repository weights, Safetensors checkpoints, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Ling-3.0-flash-VL BF16 weight release in the inclusionAI/Ling-3.0-flash-VL repository, built on the Ling-3.0-flash base. The text base and Fin continuation, the fp8, fp4 and int4 siblings, and any gateway-hosted copy are separate boundaries.", "strengths": [ "Ungated BF16 weights with SGLang launch recipes and a linked training-content summary", "Documented image and video input with a 256K context and a stated 124B total and 5.5B active parameter count" ], "tradeoffs": [ "The MIT grant rests on model-card metadata with no LICENSE file in the repository", "No first-party hosted API for the release was found in the sources read, and the card documents no release date" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.6 }, "evidence": [ { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official inclusionAI Hugging Face organization page", "url": "https://huggingface.co/inclusionAI", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-kwaipilot-kat-coder-v2-5-dev", "source_id": null, "name": "KAT-Coder-V2.5-Dev", "developer": "Kwaipilot", "url": "https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev", "description": "Kwaipilot's July 2026 open-weight coding release in the KAT line, derived from Qwen3.6-35B-A3B for agentic repository work with 256K context, published under Apache 2.0 with weights on Kwaipilot's Hugging Face organization alongside a technical report. The flagship KAT-Coder stays API-only on StreamLake. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "KAT-Coder-V2.5-Dev", "description": "Open-weight agentic coding release derived from Qwen3.6-35B-A3B, serving 256K context through vLLM on 8 GPUs", "family": "kat", "release_date": "2026-07-23", "last_updated": "2026-07-23", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The Kwaipilot/KAT-Coder-V2.5-Dev repository declares license apache-2.0 in its model metadata and names the Apache-compatible Qwen3.6-35B-A3B base. The repository ships no LICENSE file of its own, so the grant is the standard Apache 2.0 text. No separate model-use terms were found on the pages read.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "KAT-Coder-V2.5-Dev weights published in Kwaipilot's Hugging Face repository under the Apache 2.0 licence declared by the model card", "kind": "web_terms", "url": "https://opensource.org/licenses/Apache-2.0", "verified_at": "2026-10-14" } ], "access_boundary": "This record represents the KAT-Coder-V2.5-Dev open-weight checkpoint. The API-only flagship KAT-Coder on StreamLake, the earlier KAT-Dev line, the base Qwen3.6-35B-A3B weights, and Kwaipilot's HiPO, OASIS, and SRPO lines are separate boundaries.", "strengths": [ "Apache 2.0 open weights on the developer's own organization with vLLM and transformers serving commands in the model card", "Agentic coding release with a technical report, benchmark methodology, and 256K context for repository-scale work" ], "tradeoffs": [ "The card's reference serving setup needs tensor parallelism across 8 GPUs, so running it is data-center-class self-operation", "The flagship KAT-Coder stays API-only, so this Dev checkpoint trails it and its relative standing needs the report, not the card" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 4, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.12 }, "evidence": [ { "kind": "web", "label": "Kwaipilot KAT-Coder-V2.5-Dev weight repository: apache-2.0 licence, Qwen3.6 base", "url": "https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev", "verified_at": "2026-10-14" }, { "kind": "web", "label": "KAT project page: open-source Dev release and API-only flagship split", "url": "https://kwaipilot.github.io/KAT-Coder/", "verified_at": "2026-10-14" }, { "kind": "web", "label": "KAT-Coder-V2.5 Technical Report on arXiv", "url": "https://arxiv.org/abs/2607.05471", "verified_at": "2026-10-14" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev", "verified_at": "2026-10-14", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-14", "verified_at": "2026-10-14" }, { "id": "model-liquidai-lfm2-5-2-6b", "source_id": null, "name": "LFM2.5-2.6B", "developer": "Liquid AI", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B", "description": "Liquid AI's 2.6B-parameter hybrid LFM2.5 language model post-trained for agentic tool use with a 131,072-token context window, published as ungated weights with official GGUF, MLX, and ONNX exports under the LFM Open License v1.0 and served to some users through OpenRouter rather than a Liquid AI API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "LFM2.5-2.6B", "description": "LFM2.5-2.6B is part of LFM2.5, a family of hybrid models designed for on-device deployment. It builds on the LFM2 architecture with a 128K context window and agentic post-training.", "family": "lfm2.5", "release_date": "2026-08-04", "last_updated": "2026-08-24", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "other", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B" } ] }, "licenses": [ "LicenseRef-LFM-Open-License-1.0" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository LICENSE file is the LFM Open License v1.0, which reuses the Apache License 2.0 text and adds Section 5: the copyright and patent grants are conditioned on a commercial use limitation, and a legal entity with annual revenue of $10,000,000 or more has no licensed right to use the work or a derivative for commercial purposes until it buys a commercial licence from Liquid AI, Inc. Section 5(c) exempts qualified non-profit organizations from the threshold for non-commercial or research purposes, and there is no copyleft, so proprietary fine-tunes are permitted. A revenue ceiling on an otherwise Apache-2.0 grant is a use restriction, so the release is source-available rather than open source. The same terms govern the official GGUF, MLX, and ONNX exports; no separate code licence covers this repository, which ships weights, a chat template, and tokenizer files only.", "license_evidence": [ { "kind": "web_terms", "license_id": "LicenseRef-LFM-Open-License-1.0", "scope": "LiquidAI/LFM2.5-2.6B weights, chat template, and tokenizer files, and every official format export and derivative of them; Section 5 caps free commercial use at legal entities below $10,000,000 annual revenue", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/blob/main/LICENSE", "verified_at": "2026-09-28" }, { "kind": "web_terms", "license_id": "LicenseRef-LFM-Open-License-1.0", "scope": "Liquid AI's own licence page carrying the same LFM Open License v1.0 text, its summary, its hosting-platform FAQ, and its Apache 2.0 comparison", "url": "https://www.liquid.ai/lfm-license", "verified_at": "2026-09-28" } ], "access_boundary": "This record represents the LiquidAI/LFM2.5-2.6B weight release announced 4 August 2026, the 2.6B dense text-to-text checkpoint of the LFM2.5 line. The LFM2.5-2.6B-Base pre-trained checkpoint, the deprecated LFM2-2.6B and LFM2-2.6B-Exp predecessors, the sibling LFM2.5 sizes and the LFM2.5-VL, LFM2.5-Audio, and Liquid Nanos families, the official GGUF, MLX, and ONNX format exports, the LFM2.5-2.6B-DSpark speculative-decoding drafter, the Liquid Playground and the partner API programme, and OpenRouter's hosted route for the same weights are separate boundaries.", "strengths": [ "Developer-published BF16 safetensors checkpoint with official GGUF, MLX, and ONNX exports and a 328M DSpark speculative-decoding drafter, all on the LiquidAI Hugging Face organization", "The LFM Open License is published in full on the developer's own site with a plain-language summary, a hosting-platform FAQ, and an explicit Apache 2.0 comparison, and the FAQ confirms the $10 million threshold and the non-profit exemption", "First-party model card and documentation cover the hybrid convolution and attention layer layout, parameter count, 34T-token training budget, 128K mid-training context extension, the four post-training stages, the ChatML-like chat template, tool-call syntax, and per-runtime deployment recipes from llama.cpp and MLX through vLLM and SGLang" ], "tradeoffs": [ "The LFM Open License ends the right to use the release for commercial purposes once a legal entity's annual revenue reaches $10 million, so a commercial licence from Liquid AI replaces the free grant at that point", "Liquid AI states it does not offer a hosted API of its own; the only documented managed route is OpenRouter, and the Liquid Playground is a rate-limited trial", "The deprecations page names successors for retired LFM2 checkpoints and the LFM2.5 line, but the model card publishes no versioning or deprecation commitment for the 2.6B line itself, and the LFM2.5-Encoders and Liquid Nanos families are listed in the same documentation matrix" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 10, "deployment_portability": 9, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.72 }, "evidence": [ { "kind": "web", "label": "Official LFM2.5-2.6B weights repository and model card", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B", "verified_at": "2026-09-28" }, { "kind": "web", "label": "LFM Open License v1.0 file in the weights repository", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/blob/main/LICENSE", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Liquid AI LFM License page: licence text, summary, hosting FAQ, and Apache 2.0 comparison", "url": "https://www.liquid.ai/lfm-license", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official release announcement: LFM2.5-2.6B, training recipe, and on-device agent deployment", "url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Liquid documentation model matrix: HF, GGUF, MLX, and ONNX availability per model", "url": "https://docs.liquid.ai/lfm/models/complete-library", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official documentation deprecations page naming LFM2.5-2.6B as the replacement for the LFM2-2.6B and LFM2-2.6B-Exp lines", "url": "https://docs.liquid.ai/lfm/help/deprecations", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Liquid AI FAQ: no hosted API of its own, a rate-limited playground, and OpenRouter as the documented managed route", "url": "https://www.liquid.ai/faq", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Liquid Foundation Models page stating that every LFM is free to download, run, and fine-tune until the company passes $10M in annual revenue", "url": "https://www.liquid.ai/models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Model configuration giving the 131,072-token maximum position embeddings and the hybrid convolution and full-attention layer layout", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/raw/main/config.json", "verified_at": "2026-09-28" }, { "kind": "web", "label": "OpenRouter model page documenting a hosted route for the same release", "url": "https://openrouter.ai/liquid/lfm-2.5-2.6b", "verified_at": "2026-09-28" } ], "metadata_verified_at": "2026-09-28", "verified_at": "2026-09-28" }, { "id": "model-meituan-longcat-2-0", "source_id": "meituan/longcat-2.0", "name": "LongCat-2.0", "developer": "Meituan", "url": "https://huggingface.co/meituan-longcat/LongCat-2.0", "description": "Meituan's LongCat-2.0, a 1.6-trillion-parameter MoE release with 48B active parameters, LSA sparse attention, n-gram embeddings, and 1M context, published as open weights under MIT.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "LongCat-2.0", "description": "Meituan LongCat-2.0, a reasoning model with tool calling and a 1M-token context window", "family": "longcat", "release_date": "2026-06-30", "last_updated": "2026-06-30", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The LongCat-2.0 repository carries an MIT license file (Copyright (c) 2026 Meituan) at the repository root, confirming the open-weights grant. models.dev's row still reports open weights as false and attributes the org as meituan; the official publisher org is meituan-longcat.", "license_evidence": [ { "license_id": "MIT", "scope": "LongCat-2.0 repository, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/meituan-longcat/LongCat-2.0/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the LongCat-2.0 weight release published by the official meituan-longcat organization. The longcat.ai hosted chat and independent hosts are separate boundaries; models.dev attributes this row to a stale meituan org.", "strengths": [ "Frontier-scale open MoE release with an explicit MIT license file at the repository root", "Strong coding and agentic positioning with 1M context and published technical material" ], "tradeoffs": [ "The 1.8-trillion-parameter checkpoint set is impractical for most self-operated hardware", "Community support is primarily GGUF-quantized rather than first-party optimized formats" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.4 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meituan/longcat-2.0.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/meituan-longcat/LongCat-2.0", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-meituan-longcat-2-5-preview", "source_id": "meituan/longcat-2.5-preview", "name": "LongCat-2.5-Preview", "developer": "Meituan", "url": "https://longcat.chat/platform/docs/", "description": "Meituan's LongCat-2.5-Preview, a multimodal agentic model with image and video understanding and a 1M-token context window, served through the LongCat API Platform under Meituan's platform terms, with no published weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "LongCat-2.5-Preview", "description": "Meituan's multimodal reasoning model for coding and agent tasks, with image understanding and a 1M-token context window", "family": "longcat", "release_date": "2026-09-25", "last_updated": "2026-09-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-LongCat-Platform-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "LongCat-2.5-Preview is served through the LongCat API Platform under the LongCat AI Platform Terms of Use, which form part of the LongCat Terms of Use (version 3.1, updated 2026-06-30). Those terms reserve all rights in the models, including model weights and parameters, to the operator, prohibit reverse engineering the platform's models and training or fine-tuning competing models on outputs, and license no weights for self-operation. The LongCat-2.0 MIT weights license covers the separate LongCat-2.0 release and does not extend to this preview.", "license_evidence": [ { "license_id": "LicenseRef-LongCat-Platform-Terms", "scope": "Access to LongCat-2.5-Preview through the LongCat API Platform under the LongCat AI Platform Terms of Use and the LongCat Terms of Use", "kind": "web_terms", "url": "https://longcat.chat/platform/private/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the LongCat-2.5-Preview release announced September 25, 2026 and served through the LongCat API Platform under the model name LongCat-2.5-Preview. The open-weight LongCat-2.0 release, the retired LongCat-2.0-Preview beta, earlier LongCat-Flash models, the longcat.chat consumer chat product, and independent hosts are separate boundaries.", "strengths": [ "The platform documents the model name, the 1M-token context with 128K output, image and video input, OpenAI-compatible and Anthropic-compatible endpoints, and a dated changelog entry", "The governing platform terms and the parent terms of use are published with a version number and an update date" ], "tradeoffs": [ "No weights, repository, or model card are published for this preview, so self-operation is unavailable even though the LongCat-2.0 release is open", "The Preview name carries no stated retention posture, and the platform's change log records retiring six earlier LongCat models on May 29, 2026" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/meituan/longcat-2.5-preview.toml", "verified_at": "2026-10-06" }, { "kind": "web", "label": "LongCat API Platform quick start: supported models, endpoints, and context limits", "url": "https://longcat.chat/platform/docs/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "LongCat API Platform change log: LongCat-2.5-Preview release entry of September 25, 2026", "url": "https://longcat.chat/platform/docs/change-log", "verified_at": "2026-10-06" }, { "kind": "web", "label": "LongCat API visual understanding guide for LongCat-2.5-Preview", "url": "https://longcat.chat/platform/docs/image-video-understanding", "verified_at": "2026-10-06" }, { "kind": "web", "label": "LongCat AI Platform Terms of Use", "url": "https://longcat.chat/platform/private/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official meituan-longcat Hugging Face organization: no LongCat-2.5 weight repository", "url": "https://huggingface.co/meituan-longcat", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-meta-llama-3-2-11b-vision-instruct", "source_id": "meta/llama-3.2-11b-vision-instruct", "name": "Llama 3.2 11B Vision Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct", "description": "A downloadable image-and-text instruction model governed by the Llama 3.2 Community License and its incorporated acceptable-use and regional restrictions.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Llama-3.2-11B-Vision-Instruct", "description": "Open multimodal Llama model for image understanding, captioning, and visual QA", "family": "llama", "release_date": "2024-09-25", "last_updated": "2024-09-25", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-3.2-Community", "LicenseRef-Llama-3.2-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the model weights under the custom Llama 3.2 Community License, which carries attribution and redistribution duties, a large-platform commercial threshold, and an incorporated acceptable-use policy. The multimodal grant also excludes organizations established in the European Union, so public weights do not make this an open-source distribution.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.2-Community", "scope": "Llama 3.2 11B Vision Instruct weights, tokenizer, model code, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct/blob/9eb2daaa8597bf192a8b0e73f848f3a102794df5/LICENSE.txt", "verified_at": "2026-09-04" }, { "license_id": "LicenseRef-Llama-3.2-AUP", "scope": "Acceptable-use and multimodal regional restrictions incorporated into the Llama 3.2 Community License", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct/blob/9eb2daaa8597bf192a8b0e73f848f3a102794df5/USE_POLICY.md", "verified_at": "2026-09-04" } ], "access_boundary": "This record represents the Llama 3.2 11B Vision Instruct release. The 11B base model, 90B variants, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries.", "strengths": [ "Meta publishes both Transformers-compatible and original checkpoint artifacts", "The official model card documents several self-operated runtimes and hosted deployment paths" ], "tradeoffs": [ "Artifact access is gated and the custom license carries use, attribution, scale, and multimodal regional restrictions", "Runtime and hosted-provider coverage can change independently of this fixed model release" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.04 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and deployment guidance", "url": "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/ef71a4865a5821d8cd24f171a0100156a02e808e/models/meta/llama-3.2-11b-vision-instruct.toml", "verified_at": "2026-09-04" } ], "metadata_verified_at": "2026-09-04", "verified_at": "2026-09-04" }, { "id": "model-meta-llama-4-maverick-17b-instruct", "source_id": "meta/llama-4-maverick-17b-instruct", "name": "Llama 4 Maverick 17B Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "description": "A gated downloadable image-and-text instruction model governed by the Llama 4 Community License and an incorporated acceptable-use policy that withholds the multimodal grant from European Union entities.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama 4 Maverick 17B Instruct", "description": "Open multimodal Llama for strong reasoning with efficient everyday serving", "family": "llama", "release_date": "2025-04-05", "last_updated": "2025-04-05", "knowledge_cutoff": "2024-08", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-4-Community", "LicenseRef-Llama-4-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights under the custom Llama 4 Community License, which carries attribution and redistribution duties, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. The policy, not the license text, states that the Section 1(a) rights for multimodal Llama 4 models are not granted to individuals domiciled in or companies based in the European Union, so public weights do not make this an open-source distribution.", "license_evidence": [ { "license_id": "LicenseRef-Llama-4-Community", "scope": "Llama 4 Maverick 17B 128E Instruct weights, tokenizer, processor configuration, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct/blob/73d14711bcc77c16df3470856949c3764056b617/LICENSE", "verified_at": "2026-09-11" }, { "license_id": "LicenseRef-Llama-4-AUP", "scope": "Acceptable-use policy and European Union multimodal exclusion incorporated by reference into the Llama 4 Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama4/use-policy/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the meta-llama/Llama-4-Maverick-17B-128E-Instruct BF16 release dated April 5, 2025. The Maverick base model, the separately published FP8 checkpoint, Llama 4 Scout, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries.", "strengths": [ "Meta publishes Transformers-compatible Safetensors shards with tokenizer, processor, and chat-template files in one pinned repository", "The model card names the governing license, the base model, the release date, and distinguishes the BF16 release from the FP8 and Scout variants" ], "tradeoffs": [ "Artifact access is gated behind identity disclosure and the incorporated policy withholds the multimodal grant from European Union entities", "The model card documents a single Transformers path and no hosted serving paths, so other runtimes and operators need independent verification" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 6, "deployment_portability": 6, "serving_reach": 4, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 5.78 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and Transformers guidance", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Llama 4 Community License Agreement page", "url": "https://developer.meta.com/ai/llama4/license/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/meta/llama-4-maverick-17b-instruct.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-meta-llama-4-scout-17b-instruct", "source_id": "meta/llama-4-scout-17b-instruct", "name": "Llama 4 Scout 17B Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "description": "A gated downloadable image-and-text instruction model governed by the Llama 4 Community License and an incorporated acceptable-use policy that withholds the multimodal grant from European Union entities.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama 4 Scout 17B Instruct", "description": "Open Llama with long-context vision for efficient multimodal agents", "family": "llama", "release_date": "2025-04-05", "last_updated": "2025-04-05", "knowledge_cutoff": "2024-08", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 3500000, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-4-Community", "LicenseRef-Llama-4-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights under the custom Llama 4 Community License, which carries attribution and redistribution duties, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. The policy states that the rights for multimodal Llama 4 models are not granted to individuals domiciled in or companies based in the European Union, so public weights do not make this an open-source distribution. Identical terms to the reviewed Maverick record.", "license_evidence": [ { "license_id": "LicenseRef-Llama-4-Community", "scope": "Llama 4 Scout 17B 16E Instruct weights, tokenizer, processor configuration, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct/blob/92f3b1597a195b523d8d9e5700e57e4fbb8f20d3/LICENSE", "verified_at": "2026-09-11" }, { "license_id": "LicenseRef-Llama-4-AUP", "scope": "Acceptable-use policy and European Union multimodal exclusion incorporated by reference into the Llama 4 Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama4/use-policy/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the meta-llama/Llama-4-Scout-17B-16E-Instruct BF16 release dated April 5, 2025 (17B active, 109B total MoE, 10M context). The Scout base model, Llama 4 Maverick, the FP8 Maverick checkpoint, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries.", "strengths": [ "Meta publishes Transformers-compatible Safetensors shards with tokenizer, processor, and chat-template files in one pinned repository", "The model card names the governing license, the base model, the release date, and distinguishes the Scout BF16 release from the Maverick and FP8 variants" ], "tradeoffs": [ "Artifact access is gated behind identity disclosure and the incorporated policy withholds the multimodal grant from European Union entities", "The model card documents a single Transformers path and no hosted serving paths" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 6, "deployment_portability": 6, "serving_reach": 4, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 5.78 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and Transformers guidance", "url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Llama 4 Community License Agreement page", "url": "https://developer.meta.com/ai/llama4/license/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Llama 4 acceptable-use policy", "url": "https://developer.meta.com/ai/llama4/use-policy/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/meta/llama-4-scout-17b-instruct.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-meta-llama-3-1-70b-instruct", "source_id": "meta/llama-3.1-70b-instruct", "name": "Llama-3.1-70B-Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct", "description": "A gated downloadable 70B multilingual instruction model governed by the Llama 3.1 Community License and its incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-3.1-70B-Instruct", "description": "Open Llama instruction model for multilingual chat, reasoning, and coding", "family": "llama", "release_date": "2024-07-23", "last_updated": "2024-07-23", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-3.1-Community", "LicenseRef-Llama-3.1-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.1 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy, making the release source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.1-Community", "scope": "Llama-3.1-70B-Instruct weights, tokenizer, original checkpoints, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct/blob/1605565b47bb9346c5515c34102e054115b4f98b/LICENSE", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Llama-3.1-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the meta-llama/Llama-3.1-70B-Instruct release. The base model it was derived from, sibling sizes, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries. The 3.1 base model, the 8B and 405B siblings, Llama Guard 3, and Prompt Guard remain separate records.", "strengths": [ "Meta publishes both Transformers-compatible Safetensors shards and original consolidated checkpoints with a checklist file", "The model card names the governing license, the base model, a release date, and a static-model status statement" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no hosted serving paths, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.64 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.1 Community License Agreement page", "url": "https://developer.meta.com/ai/llama3_1/license/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.1 acceptable-use policy page", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/llama-3.1-70b-instruct.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-llama-3-1-8b-instruct", "source_id": "meta/llama-3.1-8b-instruct", "name": "Llama-3.1-8B-Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "description": "A gated downloadable 8B multilingual instruction model governed by the Llama 3.1 Community License and its incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-3.1-8B-Instruct", "description": "Compact open Llama model for lightweight chat, drafting, and self-hosting", "family": "llama", "release_date": "2024-07-23", "last_updated": "2024-07-23", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-3.1-Community", "LicenseRef-Llama-3.1-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.1 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy, making the release source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.1-Community", "scope": "Llama-3.1-8B-Instruct weights, tokenizer, original checkpoints, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct/blob/0e9e39f249a16976918f6564b8830bc894c89659/LICENSE", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Llama-3.1-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the meta-llama/Llama-3.1-8B-Instruct release. The base model it was derived from, sibling sizes, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries. The 3.1 base model, the 70B and 405B siblings, Llama Guard 3, and Prompt Guard remain separate records.", "strengths": [ "Meta publishes both Transformers-compatible Safetensors shards and original consolidated checkpoints with a checklist file", "The model card names the governing license, the base model, a release date, and a static-model status statement" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no hosted serving paths, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.84 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.1 Community License Agreement page", "url": "https://developer.meta.com/ai/llama3_1/license/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.1 acceptable-use policy page", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/llama-3.1-8b-instruct.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-llama-3-2-1b", "source_id": "meta/llama-3.2-1b", "name": "Llama-3.2-1B", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B", "description": "A gated downloadable 1B base text model governed by the Llama 3.2 Community License and its incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-3.2-1B", "description": "Compact open Llama base model for lightweight and on-device use", "family": "llama", "release_date": "2024-09-25", "last_updated": "2024-09-25", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Llama 3.2 Community License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B" } ] }, "licenses": [ "LicenseRef-Llama-3.2-Community", "LicenseRef-Llama-3.2-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.2 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a \"Llama\" naming requirement for derived models, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. The policy's European Union exclusion applies only to multimodal Llama 3.2 models, so it does not bind this text-only release, but the use-based terms still make it source-available.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.2-Community", "scope": "Llama-3.2-1B weights, tokenizer, original checkpoints, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B/blob/4e20de362430cd3b72f300e6b0f18e50e7166e08/LICENSE", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Llama-3.2-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama3_2/use-policy/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the meta-llama/Llama-3.2-1B release. The base model it was derived from, sibling sizes, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries. The Instruct fine-tune, the 3B sibling, and quantized variants remain separate records.", "strengths": [ "Meta publishes both Transformers-compatible Safetensors shards and original consolidated checkpoints with a checklist file", "The model card names the governing license, the base model, a release date, and a static-model status statement" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no hosted serving paths, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.2 Community License Agreement page", "url": "https://developer.meta.com/ai/llama3_2/license/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.2 acceptable-use policy page", "url": "https://developer.meta.com/ai/llama3_2/use-policy/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/llama-3.2-1b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-llama-3-2-1b-instruct", "source_id": null, "name": "Llama-3.2-1B-Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct", "description": "A gated downloadable 1B instruction-tuned text model governed by the Llama 3.2 Community License and its incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-3.2-1B-Instruct", "description": "Llama 3.2 1B instruction-tuned multilingual text model for dialogue, agentic retrieval and summarization", "family": "llama", "release_date": "2024-09-25", "last_updated": null, "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 131072, "input": null, "output": null }, "reported_open_weights": null, "reported_license": "Llama 3.2 Community License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-3.2-Community", "LicenseRef-Llama-3.2-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.2 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a \"Llama\" naming requirement for derived models, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. The policy's European Union exclusion applies only to multimodal Llama 3.2 models, so it does not bind this text-only release, but the use-based terms still make it source-available.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.2-Community", "scope": "Llama-3.2-1B-Instruct weights, tokenizer, original checkpoints, and related Llama materials, reproduced in the gated model page", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-Llama-3.2-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama 3.2 Community License and reproduced in the gated model page", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the meta-llama/Llama-3.2-1B-Instruct release of September 25, 2024 at Hugging Face commit 9213176726f574b556790deb65791e0c5aa438b6. The base 1B model, the sibling size, the quantized SpinQuant and QLoRA variants Meta describes in the same card, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries.", "strengths": [ "Meta publishes Transformers-compatible Safetensors weights and an original-format checkpoint in the same repository (model.safetensors file plus original consolidated checkpoint)", "The model card names the governing license, a release date, and a static-model status statement" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no hosted serving paths, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card, license text, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official repository metadata listing LICENSE.txt, USE_POLICY.md, and checkpoint files", "url": "https://huggingface.co/api/models/meta-llama/Llama-3.2-1B-Instruct", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-meta-llama-3-2-3b", "source_id": "meta/llama-3.2-3b", "name": "Llama-3.2-3B", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B", "description": "A gated downloadable 3B base text model governed by the Llama 3.2 Community License and its incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-3.2-3B", "description": "Small open Llama base model for lightweight text generation and self-hosting", "family": "llama", "release_date": "2024-09-25", "last_updated": "2024-09-25", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Llama 3.2 Community License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B" } ] }, "licenses": [ "LicenseRef-Llama-3.2-Community", "LicenseRef-Llama-3.2-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.2 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a \"Llama\" naming requirement for derived models, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. The policy's European Union exclusion applies only to multimodal Llama 3.2 models, so it does not bind this text-only release, but the use-based terms still make it source-available.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.2-Community", "scope": "Llama-3.2-3B weights, tokenizer, original checkpoints, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B/blob/13afe5124825b4f3751f836b40dafda64c1ed062/LICENSE", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Llama-3.2-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama3_2/use-policy/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the meta-llama/Llama-3.2-3B release. The base model it was derived from, sibling sizes, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries. The Instruct fine-tune, the 1B sibling, and quantized variants remain separate records.", "strengths": [ "Meta publishes both Transformers-compatible Safetensors shards and original consolidated checkpoints with a checklist file", "The model card names the governing license, the base model, a release date, and a static-model status statement" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no hosted serving paths, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.2 Community License Agreement page", "url": "https://developer.meta.com/ai/llama3_2/license/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.2 acceptable-use policy page", "url": "https://developer.meta.com/ai/llama3_2/use-policy/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/llama-3.2-3b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-llama-3-2-3b-instruct", "source_id": null, "name": "Llama-3.2-3B-Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "description": "A gated downloadable 3B instruction-tuned text model governed by the Llama 3.2 Community License and its incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-3.2-3B-Instruct", "description": "Llama 3.2 3B instruction-tuned multilingual text model for dialogue, agentic retrieval and summarization", "family": "llama", "release_date": "2024-09-25", "last_updated": null, "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 131072, "input": null, "output": null }, "reported_open_weights": null, "reported_license": "Llama 3.2 Community License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-3.2-Community", "LicenseRef-Llama-3.2-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.2 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a \"Llama\" naming requirement for derived models, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. The policy's European Union exclusion applies only to multimodal Llama 3.2 models, so it does not bind this text-only release, but the use-based terms still make it source-available.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.2-Community", "scope": "Llama-3.2-3B-Instruct weights, tokenizer, original checkpoints, and related Llama materials, reproduced in the gated model page", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-Llama-3.2-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama 3.2 Community License and reproduced in the gated model page", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the meta-llama/Llama-3.2-3B-Instruct release of September 25, 2024 at Hugging Face commit 0cb88a4f764b7a12671c53f0838cd831a0843b95. The base 3B model, the sibling size, the quantized SpinQuant and QLoRA variants Meta describes in the same card, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries.", "strengths": [ "Meta publishes Transformers-compatible Safetensors weights and an original-format checkpoint in the same repository (two Safetensors shards plus original consolidated checkpoint)", "The model card names the governing license, a release date, and a static-model status statement" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no hosted serving paths, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card, license text, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official repository metadata listing LICENSE.txt, USE_POLICY.md, and checkpoint files", "url": "https://huggingface.co/api/models/meta-llama/Llama-3.2-3B-Instruct", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-meta-llama-3-3-70b-instruct", "source_id": "meta/llama-3.3-70b-instruct", "name": "Llama-3.3-70B-Instruct", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct", "description": "A gated downloadable text instruction model governed by the Llama 3.3 Community License and its incorporated acceptable-use policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-3.3-70B-Instruct", "description": "Popular open Llama workhorse for multilingual chat, coding, and self-hosting", "family": "llama", "release_date": "2024-12-06", "last_updated": "2024-12-06", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct" } ] }, "licenses": [ "LicenseRef-Llama-3.3-Community", "LicenseRef-Llama-3.3-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.3 Community License, which carries attribution and redistribution duties, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. The policy's European Union exclusion applies only to multimodal Llama 3.3 models, so it does not bind this text-only release, but the use-based terms still make it source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.3-Community", "scope": "Llama 3.3 70B Instruct weights, tokenizer, original checkpoints, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct/blob/6f6073b423013f6a7d4d9f39144961bfbfbc386b/LICENSE", "verified_at": "2026-09-11" }, { "license_id": "LicenseRef-Llama-3.3-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama 3.3 Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama3_3/use-policy/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the meta-llama/Llama-3.3-70B-Instruct release dated December 6, 2024. The Llama 3.1 70B base model it was tuned from, quantizations, fine-tunes, Meta-hosted products, and independent serving providers are separate model or service boundaries.", "strengths": [ "Meta publishes both Transformers-compatible Safetensors shards and original consolidated checkpoints with a checklist file", "The model card names the governing license, the base model, a release date, and a static-model status statement" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no hosted serving paths, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.64 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Llama 3.3 Community License Agreement page", "url": "https://developer.meta.com/ai/llama3_3/license/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/meta/llama-3.3-70b-instruct.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-meta-llama-guard-3-8b", "source_id": "meta/llama-guard-3-8b", "name": "Llama-Guard-3-8B", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-Guard-3-8B", "description": "A gated downloadable 8B safety classifier fine-tuned from Llama 3.1 8B that emits safe/unsafe verdicts as generated text, governed by the Llama 3.1 Community License.", "model_type": "judgment_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-Guard-3-8B", "description": "Llama 3.1-based safety classifier for moderating prompts and model responses", "family": "llama", "release_date": "2024-07-23", "last_updated": "2024-07-23", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-Guard-3-8B" } ] }, "licenses": [ "LicenseRef-Llama-3.1-Community", "LicenseRef-Llama-3.1-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights, tokenizer, and original checkpoints under the custom Llama 3.1 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy, making the release source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.1-Community", "scope": "Llama-Guard-3-8B weights, tokenizer, and related Llama materials", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-Guard-3-8B/blob/7327bd9f6efbbe6101dc6cc4736302b3cbb6e425/LICENSE", "verified_at": "2026-09-18" }, { "license_id": "LicenseRef-Llama-3.1-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama 3.1 Community License", "kind": "web_terms", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the meta-llama/Llama-Guard-3-8B release. The 1B and 11B-Vision Guard siblings, the INT8 precision sibling, the Llama 3.1 base model, quantizations, fine-tunes, and independent serving providers are separate model or service boundaries.", "strengths": [ "The model card confirms it generates text verdicts as an autoregressive language model and documents the S1-S14 hazard taxonomy", "Meta publishes Transformers-compatible Safetensors shards, tokenizer, and original checkpoints" ], "tradeoffs": [ "Artifact access is gated behind a contact-information agreement and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card documents no first-party hosted path, so independent availability must be verified with each operator" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.94 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-Guard-3-8B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.1 Community License Agreement page", "url": "https://developer.meta.com/ai/llama3_1/license/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Llama 3.1 acceptable-use policy page", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/llama-guard-3-8b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-llama-guard-4-12b", "source_id": null, "name": "Llama-Guard-4-12B", "developer": "Meta", "url": "https://huggingface.co/meta-llama/Llama-Guard-4-12B", "description": "A gated downloadable 12B natively multimodal safety classifier pruned from Llama 4 Scout that emits safe/unsafe verdicts as generated text, governed by the Llama 4 Community License.", "model_type": "judgment_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama-Guard-4-12B", "description": "Natively multimodal 12B safety classifier for text and image prompts and responses, aligned to the MLCommons hazards taxonomy", "family": "llama", "release_date": null, "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": null, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Llama 4 Community License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-Guard-4-12B" } ] }, "licenses": [ "LicenseRef-Llama-4-Community", "LicenseRef-Llama-4-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Meta publishes the weights under the custom Llama 4 Community License, which carries attribution and redistribution duties, a \"Built with Llama\" display requirement, a 700 million monthly-active-user commercial threshold, and an incorporated acceptable-use policy. That policy withholds the Section 1(a) grant from European Union entities for multimodal models included in Llama 4; this card does not say whether Llama Guard 4, which accepts images, falls inside that wording, so the exclusion is recorded as a possible restriction. The terms make the release source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Llama-4-Community", "scope": "Llama-Guard-4-12B weights, tokenizer, processor, and related Llama materials, reproduced in the gated model page", "kind": "web_terms", "url": "https://huggingface.co/meta-llama/Llama-Guard-4-12B", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-Llama-4-AUP", "scope": "Acceptable-use policy and European Union multimodal exclusion incorporated by reference into the Llama 4 Community License", "kind": "web_terms", "url": "https://www.llama.com/llama4/use-policy/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the meta-llama/Llama-Guard-4-12B release at Hugging Face commit 87acb4b94e930c3d679e6e7ee9d57e2feab9ea71. The earlier Llama Guard 3 releases, the Llama 4 Scout base it was pruned from, quantizations, fine-tunes, the Llama Moderations API, and independent serving providers are separate model or service boundaries. The model acts as a classifier but generates text verdicts, so it is recorded as a multimodal language model with the safety-classifier role carried in prose.", "strengths": [ "The model card confirms the model generates text verdicts as an autoregressive LLM and documents its hazard taxonomy, architecture, and pruning method", "Meta publishes Transformers-compatible Safetensors shards, tokenizer, and processor files" ], "tradeoffs": [ "Artifact access is gated behind a manual approval and the custom license carries attribution, scale, and acceptable-use restrictions", "The model card gives no release date or static-model statement, and its only serving mention is integration into Meta's Llama Moderations API" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 4, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 5.9 }, "evidence": [ { "kind": "web", "label": "Official model card, license text, and runtime guidance", "url": "https://huggingface.co/meta-llama/Llama-Guard-4-12B", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Llama 4 acceptable-use policy page", "url": "https://www.llama.com/llama4/use-policy/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official repository metadata listing LICENSE, USE_POLICY.md, and Safetensors files", "url": "https://huggingface.co/api/models/meta-llama/Llama-Guard-4-12B", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-meta-muse-glimmer-30b", "source_id": "meta/muse-glimmer-30b", "name": "Muse Glimmer 30B", "developer": "Meta", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "description": "An open-weights 29.6B text-and-image model distilled from Muse Spark, released under Apache-2.0 with a frozen perception encoder, a 131,072-token context, and an advisory usage policy.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Muse Glimmer 30B", "description": "Muse Glimmer is a 30-billion-parameter open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark for always-on local agents, tool use, coding, and image understanding.", "family": "muse", "release_date": "2026-08-10", "last_updated": "2026-08-10", "knowledge_cutoff": "2026-01-04", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [ { "label": "Announcement", "url": "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", "type": "announcement" }, { "label": "Model card", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "type": "model_card" }, { "label": "Developer docs", "url": "https://developer.meta.com/ai/models/muse-glimmer/", "type": "docs" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships an unmodified Apache License 2.0 text at the pinned commit, and the card states \"All artifacts are released under Apache 2.0\". A separate USAGE_POLICY.md adds age and prohibited-use duties but no use restriction.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Muse-Glimmer-30B weights, tokenizer, chat template, and related artifacts in the official meta-models repository", "kind": "web_terms", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B/blob/a4e59da52a7bc87ae7251dd5545c0dd437c44b68/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the meta-models/Muse-Glimmer-30B release of August 2026 at Hugging Face commit a4e59da52a7bc87ae7251dd5545c0dd437c44b68. The GGUF and ExecuTorch companion artifacts, the DFlash drafter, the perception encoder, the Muse Spark line, quantizations, fine-tunes, and independent hosting providers remain separate boundaries.", "strengths": [ "Ungated Apache-2.0 weights with the full license text at the pinned commit", "Dated release with a model card, research announcement, developer documentation, and wide runtime coverage" ], "tradeoffs": [ "A separate acceptable-use policy adds age and prohibited-use duties", "No successor or deprecation posture is published yet for the line" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.74 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and license text", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official release announcement", "url": "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official developer documentation", "url": "https://dev.meta.ai/docs/muse-glimmer", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/muse-glimmer-30b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-muse-spark-1-1", "source_id": "meta/muse-spark-1.1", "name": "Muse Spark 1.1", "developer": "Meta", "url": "https://research.meta.ai/blog/introducing-muse-spark-meta-model-api", "description": "An API-served multimodal release of the Muse Spark family with a 1,048,576-token context, available through the Meta Model API and Muse Code without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Muse Spark 1.1", "description": "Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.", "family": "muse", "release_date": "2026-04-08", "last_updated": "2026-07-09", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "pdf", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Meta-Model-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Meta serves the Muse Spark family through the Meta Model API under the Meta Model API Terms of Service, which incorporate the acceptable-use and geographic-use policies by reference. The terms reserve all rights in the services, prohibit reverse engineering or distillation of the underlying models, and bar using the service to train competing models; no weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Meta-Model-API-Terms", "scope": "Access to the Muse Spark checkpoint through the Meta Model API under the Terms of Service with the incorporated acceptable-use and geographic-use policies", "kind": "web_terms", "url": "https://ai.developer.meta.com/legal/terms-of-service", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the Muse Spark 1.1 named release announced July 9, 2026 and served under the Meta Model API. The base Muse Spark launch, the 1.2 and 1.3 checkpoints, Muse Code, and reseller surfaces remain separate boundaries.", "strengths": [ "Dated named release with an announcement, evaluation report, and API documentation", "Meta documents the fixed model identity and tier placement in its API terms" ], "tradeoffs": [ "No model weights are published, so the release cannot be self-operated", "Preview terms are replaceable at Meta's discretion and no retirement policy is published" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.14 }, "evidence": [ { "kind": "web", "label": "Official Muse Spark 1.1 announcement", "url": "https://research.meta.ai/blog/introducing-muse-spark-meta-model-api", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Meta Model API documentation", "url": "https://ai.developer.meta.com/docs/models", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Meta Model API terms of service", "url": "https://ai.developer.meta.com/legal/terms-of-service", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/muse-spark-1.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-muse-spark-1-2", "source_id": "meta/muse-spark-1.2", "name": "Muse Spark 1.2", "developer": "Meta", "url": "https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2", "description": "A coding-focused API-served update to Muse Spark 1.1 with a 1,048,576-token context, available through the Meta Model API and Muse Code without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Muse Spark 1.2", "description": "Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.", "family": "muse", "release_date": "2026-08-05", "last_updated": "2026-08-05", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "pdf", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Meta-Model-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Meta serves the Muse Spark family through the Meta Model API under the Meta Model API Terms of Service, which incorporate the acceptable-use and geographic-use policies by reference. The terms reserve all rights in the services, prohibit reverse engineering or distillation of the underlying models, and bar using the service to train competing models; no weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Meta-Model-API-Terms", "scope": "Access to the Muse Spark checkpoint through the Meta Model API under the Terms of Service with the incorporated acceptable-use and geographic-use policies", "kind": "web_terms", "url": "https://ai.developer.meta.com/legal/terms-of-service", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the Muse Spark 1.2 checkpoint announced August 5, 2026 and served under the Meta Model API. The 1.1 and 1.3 checkpoints, the contributor tier, Muse Code, and reseller surfaces remain separate boundaries.", "strengths": [ "Dated named checkpoint with announcement, methodology report, and expanded global API access", "Meta documents the checkpoint identity, tier placement, and multimodal coverage in its API terms" ], "tradeoffs": [ "No model weights are published, so the release cannot be self-operated", "Preview terms are replaceable at Meta's discretion and no retirement policy is published" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Muse Spark 1.2 announcement", "url": "https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Muse Spark 1.2 multimodal follow-up", "url": "https://research.meta.ai/blog/multimodal-intelligence-of-muse-spark-1-2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Meta Model API documentation", "url": "https://ai.developer.meta.com/docs/models", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Meta Model API terms of service", "url": "https://ai.developer.meta.com/legal/terms-of-service", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/muse-spark-1.2.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-meta-muse-spark-1-3", "source_id": "meta/muse-spark-1.3", "name": "Muse Spark 1.3", "developer": "Meta", "url": "https://research.meta.ai/blog/introducing-muse-spark-1-3", "description": "The newest API-served Muse Spark checkpoint with max reasoning, a 1,048,576-token context, and reduced token and tool-call usage, available through the Meta Model API and Muse Code without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Muse Spark 1.3", "description": "Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It improves long-horizon agent collaboration, instruction following, and coding efficiency relative to Muse Spark 1.2.", "family": "muse", "release_date": "2026-09-02", "last_updated": "2026-09-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "pdf", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Meta-Model-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Meta serves the Muse Spark family through the Meta Model API under the Meta Model API Terms of Service, which incorporate the acceptable-use and geographic-use policies by reference. The terms reserve all rights in the services, prohibit reverse engineering or distillation of the underlying models, and bar using the service to train competing models; no weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Meta-Model-API-Terms", "scope": "Access to the Muse Spark checkpoint through the Meta Model API under the Terms of Service with the incorporated acceptable-use and geographic-use policies", "kind": "web_terms", "url": "https://ai.developer.meta.com/legal/terms-of-service", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the Muse Spark 1.3 checkpoint announced September 2, 2026 and served under the Meta Model API. The 1.1 and 1.2 checkpoints, the contributor tier, Muse Code, and reseller surfaces remain separate boundaries.", "strengths": [ "Dated named checkpoint with announcement, reasoning capabilities, and documented token efficiency", "Meta documents the checkpoint identity and tier placement in its API terms" ], "tradeoffs": [ "No model weights are published, so the release cannot be self-operated", "Preview terms are replaceable at Meta's discretion and no retirement policy is published" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.4 }, "evidence": [ { "kind": "web", "label": "Official Muse Spark 1.3 announcement", "url": "https://research.meta.ai/blog/introducing-muse-spark-1-3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Meta Model API documentation", "url": "https://ai.developer.meta.com/docs/models", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Meta Model API terms of service", "url": "https://ai.developer.meta.com/legal/terms-of-service", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/meta/muse-spark-1.3.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-microsoft-mai-code-1-flash", "source_id": "microsoft/mai-code-1-flash", "name": "MAI-Code-1-Flash", "developer": "Microsoft", "url": "https://microsoft.ai/models/mai-code-1-flash/", "description": "Microsoft's 137-billion-parameter code model with a 5B active set and 256K context, served inside GitHub Copilot and Visual Studio Code, with a knowledge cutoff of December 2025.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "MAI-Code-1-Flash", "description": "Microsoft coding model built for fast, efficient assistance in everyday developer workflows", "family": "mai", "release_date": "2026-06-02", "last_updated": "2026-06-08", "knowledge_cutoff": "2025-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [ { "label": "Model card", "url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "type": "model_card" }, { "label": "Announcement", "url": "https://microsoft.ai/news/introducingmai-code-1-flash/", "type": "announcement" } ], "weights": [] }, "licenses": [ "LicenseRef-Microsoft-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The official model card states the model is licensed under the product and service terms where it is deployed, such as the GitHub Copilot terms; no weights are distributed, so no weight-scope license exists. The Microsoft release GitHub repository's MIT license covers repository content only and is not model evidence.", "license_evidence": [ { "license_id": "LicenseRef-Microsoft-API-Terms", "scope": "MAI-Code-1-Flash model card terms for the June 2, 2026 release; the model is served, not distributed", "kind": "web_terms", "url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the MAI-Code-1-Flash release of June 2, 2026, served only through Microsoft developer surfaces such as GitHub Copilot and Visual Studio Code. The later MAI-Code-1.1-Flash snapshot and independent hosts are separate boundaries.", "strengths": [ "First-party model card, data card, and dated release announcement establish the identity", "Documented serving path through GitHub Copilot and Visual Studio Code with dated EU availability" ], "tradeoffs": [ "Closed weights served only on Microsoft-owned surfaces with admin-gated enablement", "The public model page has been re-titled for the 1.1 successor, so the June release must be pinned to the model card" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.26 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/microsoft/mai-code-1-flash.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://microsoft.ai/models/mai-code-1-flash/", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-microsoft-mai-code-1-1-flash", "source_id": "microsoft/mai-code-1.1-flash", "name": "MAI-Code-1.1-Flash", "developer": "Microsoft", "url": "https://microsoft.ai/models/mai-code-1.1-flash/", "description": "Microsoft's August 11, 2026 MAI-Code-1.1-Flash, a vision-capable code model with 138B parameters and a 5B active set over 256K context, built on the MAI-Thinking-1 reasoning stack and served in GitHub Copilot.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "MAI-Code-1.1-Flash", "description": "Microsoft coding model with native vision support, optimized for fast and efficient software development", "family": "mai", "release_date": "2026-08-11", "last_updated": "2026-08-11", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 256000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [ { "label": "Announcement", "url": "https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/", "type": "announcement" } ], "weights": [] }, "licenses": [ "LicenseRef-Microsoft-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The official model card states the model is licensed under the product and service terms where it is deployed, including the GitHub terms for purchasers; no weights are distributed, so no weight-scope license exists. The Microsoft release GitHub repository's MIT license covers repository content only and is not model evidence.", "license_evidence": [ { "license_id": "LicenseRef-Microsoft-API-Terms", "scope": "MAI-Code-1.1-Flash model card terms for the August 11, 2026 release; the model is served, not distributed", "kind": "web_terms", "url": "https://microsoft.ai/pdf/MAI-Code-1.1-Flash-Model-Card.PDF", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the MAI-Code-1.1-Flash release of August 11, 2026, the vision-capable successor snapshot served through GitHub Copilot and Visual Studio Code. The earlier MAI-Code-1-Flash release and independent hosts are separate boundaries.", "strengths": [ "First-party model card, data card, and dated release announcement; native image input is distinct from the 1-Flash release", "Documented production serving through GitHub Copilot with dated release and EU availability" ], "tradeoffs": [ "Closed weights served only on Microsoft-owned surfaces with admin-gated enablement", "The parent model page now surfaces 1.1 as the canonical MAI-Code line, so identity is pinned to the August release and model card" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.26 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/microsoft/mai-code-1.1-flash.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://microsoft.ai/models/mai-code-1.1-flash/", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-microsoft-phi-4", "source_id": null, "name": "Phi-4", "developer": "Microsoft", "url": "https://huggingface.co/microsoft/phi-4", "description": "A downloadable 14B dense decoder-only language model published by Microsoft Research under the MIT License with Safetensors weights.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Phi-4", "description": "14B parameter dense decoder-only Transformer built on synthetic, filtered public-domain web, and academic book data, aimed at memory- and latency-constrained reasoning use", "family": "phi", "release_date": "2024-12-12", "last_updated": null, "knowledge_cutoff": "2024-06", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 16384, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MIT", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/microsoft/phi-4" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the MIT License with a Microsoft Corporation copyright line, and the model card front matter and summary table both name MIT, so the published weights and tokenizer are open source.", "license_evidence": [ { "license_id": "MIT", "scope": "microsoft/phi-4 model weights, tokenizer, and configuration", "kind": "web_terms", "url": "https://huggingface.co/microsoft/phi-4/blob/2db69c1c3e91a05d2c64a3185acfbaf36f744e25/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the microsoft/phi-4 release of December 12, 2024 at Hugging Face commit 2db69c1c3e91a05d2c64a3185acfbaf36f744e25. Phi-4-mini, Phi-4-reasoning, Phi-4-multimodal, community quantizations and conversions, and Azure-hosted deployments are separate boundaries.", "strengths": [ "Microsoft publishes ungated Safetensors weights with a repository LICENSE file and a separate data summary card", "The model card states the architecture, context length, release date, and a static-model status with a data cutoff" ], "tradeoffs": [ "The model card documents only a Transformers path and no first-party hosted serving for this model", "The model card gives a release date and a static-model statement but no changelog for later repository commits" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.26 }, "evidence": [ { "kind": "web", "label": "Official model card, architecture, release date, and Transformers guidance", "url": "https://huggingface.co/microsoft/phi-4", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official repository LICENSE file at the reviewed commit", "url": "https://huggingface.co/microsoft/phi-4/blob/2db69c1c3e91a05d2c64a3185acfbaf36f744e25/LICENSE", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-microsoft-phi-4-mini", "source_id": "microsoft/phi-4-mini", "name": "Phi-4-mini", "developer": "Microsoft", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "description": "A downloadable instruction-tuned language model published by Microsoft under the unmodified MIT License with Safetensors weights and a companion ONNX repository.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Phi-4-mini", "description": "Compact Microsoft instruction model tuned for efficient coding assistance, reasoning, and low-latency agent tasks", "family": "phi", "release_date": "2024-12-11", "last_updated": "2024-12-11", "knowledge_cutoff": "2023-10", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [ { "label": "Weights", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "type": "weights" } ], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified MIT License with a Microsoft Corporation copyright line and governs the published weights, tokenizer, and custom modeling code. The model card front matter links the same file, so the release is open source.", "license_evidence": [ { "license_id": "MIT", "scope": "microsoft/Phi-4-mini-instruct model weights, tokenizer, configuration, and custom modeling code", "kind": "web_terms", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the microsoft/Phi-4-mini-instruct weight release at Hugging Face commit cfbefacb99257ffa30c83adab238a50856ac3083, last modified 2025-12-10. The Phi-4-mini-instruct-onnx repository, Phi-4-mini-reasoning, Phi-4-multimodal-instruct, community quantizations, and Azure-hosted deployments remain separate boundaries.", "strengths": [ "Microsoft publishes ungated Safetensors weights with a repository license file and a companion ONNX repository for the same release", "The model card documents Transformers, vLLM, SGLang, and Docker Model Runner serving paths and names Azure AI Studio as a managed option" ], "tradeoffs": [ "The repository ships custom modeling code and pins specific library versions, so runtime upgrades need independent checking", "The model card gives a release month and a static-model statement but no changelog for later repository commits" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 9, "overall": 8.02 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official companion ONNX repository for the same release", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct-onnx", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/microsoft/phi-4-mini.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-minimax-image-01", "source_id": "minimax/image-01", "name": "MiniMax image-01", "developer": "MiniMax", "url": "https://platform.minimax.io/docs/guides/image-generation", "description": "MiniMax's image-01 text-to-image generation model with single-subject reference-image support, served only through the MiniMax Open Platform API under the platform terms of service, without downloadable weights.", "model_type": "image_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "MiniMax image-01", "description": "MiniMax text-to-image generation model with reference-image support", "family": "minimax", "release_date": "2025-02-15", "last_updated": "2026-08-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 0, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-MiniMax-Platform-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "image-01 is provided as a managed service under the MiniMax Open Platform Terms of Service (effective March 30, 2026), which bar reverse engineering and attempts to extract source code from the services and bar reselling or distributing the services outside integrated applications. No model weights are licensed for self-operation, and no weights repository for image-01 appears in the developer's Hugging Face organization listing.", "license_evidence": [ { "license_id": "LicenseRef-MiniMax-Platform-Terms", "scope": "Access to image-01 through the MiniMax Open Platform under the Terms of Service", "kind": "web_terms", "url": "https://www.minimax.io/platform/protocol/terms-of-service", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the image-01 image generation model released February 15, 2025 and served on the MiniMax Open Platform under the model name image-01. The MiniMax video, speech, and music models, the M-series language models, the MiniMax Agent product, and reseller surfaces remain separate boundaries.", "strengths": [ "MiniMax documents the model name, the text-to-image and single-reference request shapes, and a per-image price on first-party pages", "The release notes record a dated release entry for the model" ], "tradeoffs": [ "No weights are published for image-01, so self-operation is not available", "No versioning, retention, or deprecation posture is published for the undated model name" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/minimax/image-01.toml", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax image generation guide documenting model image-01 and reference-image input", "url": "https://platform.minimax.io/docs/guides/image-generation", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax pay-as-you-go pricing page listing image-01", "url": "https://platform.minimax.io/docs/guides/pricing-paygo", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax API docs release notes: Image-01 release entry of February 15, 2025", "url": "https://platform.minimax.io/docs/release-notes/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax Open Platform Terms of Service", "url": "https://www.minimax.io/platform/protocol/terms-of-service", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official MiniMaxAI Hugging Face organization: no image-01 weight repository", "url": "https://huggingface.co/MiniMaxAI", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-minimax-minimax-m1", "source_id": null, "name": "MiniMax-M1", "developer": "MiniMax", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k", "description": "MiniMax's open-weight hybrid-attention reasoning model, built on the 456-billion-parameter MiniMax-Text-01 mixture-of-experts base with a 1M-token native context, published in 40K and 80K thinking-budget checkpoints under Apache-2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "MiniMax-M1", "description": "MiniMax's open-weight hybrid-attention reasoning model with function calling and a 1M-token native context, released as 40K and 80K thinking-budget checkpoints", "family": "minimax", "release_date": null, "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 1000000, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "apache-2.0", "links": [ { "label": "GitHub", "url": "https://github.com/MiniMax-AI/MiniMax-M1", "type": "docs" } ], "weights": [ { "label": "Hugging Face (80K thinking budget)", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k" }, { "label": "Hugging Face (40K thinking budget)", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-40k" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "Both weight repositories (MiniMax-M1-40k and MiniMax-M1-80k) carry a standard Apache License 2.0 LICENSE file beside the weights, the model cards declare apache-2.0, and the GitHub repository carries the same Apache-2.0 text, so the release is open-licensed with no field-of-use restriction, unlike the MiniMax Model License on the preceding Text-01 release.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "MiniMax-M1 model weights and repository files in the MiniMaxAI/MiniMax-M1-80k and MiniMaxAI/MiniMax-M1-40k Hugging Face repositories and the MiniMax-AI/MiniMax-M1 repository", "kind": "web_terms", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/main/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the MiniMax-M1 reasoning release as the pair of checkpoints MiniMaxAI/MiniMax-M1-40k and MiniMaxAI/MiniMax-M1-80k, both last modified 2025-07-07. The MiniMax-Text-01 base, the transformers-native -hf conversions, the M2 and later M-series models, downstream quantizations, and the MiniMax Open Platform are separate boundaries.", "strengths": [ "Standard Apache-2.0 text beside the weights in both checkpoints and in the GitHub repository, with no field-of-use clause", "The model card and repository document vLLM and Transformers deployment guides, function calling, and a technical report" ], "tradeoffs": [ "The 456B-parameter checkpoints need multi-GPU serving, and the release ships in two thinking-budget variants that a user must choose between", "The current MiniMax platform documentation does not list M1, so no first-party hosted endpoint is documented today" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 2, "lifecycle_transparency": 4, "documentation_provenance": 8, "overall": 7.1 }, "evidence": [ { "kind": "web", "label": "Official MiniMax-M1-80k model card and checkpoint", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official MiniMax-M1-40k model card and checkpoint", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-40k", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Apache-2.0 LICENSE file in the weights repository", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/main/LICENSE", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax-M1 GitHub repository with Apache-2.0 LICENSE and deployment guides", "url": "https://github.com/MiniMax-AI/MiniMax-M1", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-minimax-minimax-m2", "source_id": "minimax/MiniMax-M2", "name": "MiniMax-M2", "developer": "MiniMax", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "description": "A developer-published reasoning language-model release whose downloadable weights are governed by a modified MIT license, hosted in the developer's GitHub repository, that adds a usage-threshold attribution clause.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiniMax-M2", "description": "Efficient open MiniMax model built for coding agents and tool-heavy workflows", "family": "minimax", "release_date": "2025-10-27", "last_updated": "2025-10-27", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 204800, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2" } ] }, "licenses": [ "LicenseRef-MiniMax-M2-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card declares a modified MIT license and links to the LICENSE file in the MiniMax-M2 GitHub repository; the Hugging Face weight repository carries no license file of its own. That document reproduces the MIT grant but requires commercial products above 100 million monthly active users or 30 million US dollars in annual recurring revenue to prominently display 'MiniMax M2', so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-MiniMax-M2-Modified-MIT", "scope": "MiniMax-M2 model weights, repository code, and derivative works, as linked from the Hugging Face model card", "kind": "web_terms", "url": "https://github.com/MiniMax-AI/MiniMax-M2/blob/a01f06af0e56b2bc992c81354155fa9c29d8a2ed/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the MiniMaxAI/MiniMax-M2 repository release at commit 757303d492a50514c312788b5247a4f696a4c6a3 (last modified 2025-12-23). MiniMax-M2.1 and later M2.x releases, the MiniMax Agent product, the MiniMax Open Platform API, and downstream quantizations remain separate boundaries.", "strengths": [ "MiniMax publishes Safetensors shards with in-repository deployment guides for vLLM, SGLang, Transformers, and MLX", "The MiniMax Open Platform documentation lists MiniMax-M2 as a served model alongside its successors" ], "tradeoffs": [ "The governing license lives in a separate GitHub repository rather than alongside the weights, and its attribution threshold makes the release source-available", "The model card carries no release date, and the platform documentation labels the model historical without a stated retirement date" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 7.04 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and in-repository deployment guides", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official MiniMax Open Platform text-generation documentation listing MiniMax-M2", "url": "https://platform.minimax.io/docs/guides/text-generation", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/minimax/MiniMax-M2.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-minimax-minimax-m2-1", "source_id": "minimax/MiniMax-M2.1", "name": "MiniMax-M2.1", "developer": "MiniMax", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1", "description": "MiniMax's 229-billion-parameter reasoning release for coding agents and tool-heavy workflows, published as downloadable weights under a modified MIT license that requires any commercial product or service to display the model name.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiniMax-M2.1", "description": "Earlier MiniMax agent model for practical coding and productivity tasks", "family": "minimax", "release_date": "2025-12-23", "last_updated": "2025-12-23", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 204800, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1" } ] }, "licenses": [ "LicenseRef-MiniMax-M2.1-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The Hugging Face weight repository carries no license file of its own and points to the LICENSE file in the MiniMax-M2.1 GitHub repository. That document reproduces the MIT grant but requires any commercial product or service using the software or its derivatives to prominently display 'MiniMax M2.1' — with no user-count or revenue threshold, unlike the M2 terms — so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-MiniMax-M2.1-Modified-MIT", "scope": "MiniMax-M2.1 model weights, repository code, and derivative works, as linked from the Hugging Face model card", "kind": "web_terms", "url": "https://github.com/MiniMax-AI/MiniMax-M2.1/blob/ea93fc40128a5bc366e3531a924f01f3d6d9cc15/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the MiniMaxAI/MiniMax-M2.1 repository release at Hugging Face commit cd97f59135f37b2a6bf09356e485d5e4aeb7dc9c (last modified 2026-02-13). The MiniMax-M2 predecessor, later M2.x releases and the M3 generation, the highspeed endpoint tier, the MiniMax Agent product, the MiniMax Open Platform API, and downstream quantizations remain separate boundaries.", "strengths": [ "MiniMax publishes 130 Safetensors shards with in-repository deployment guides for SGLang, vLLM, Transformers, and MLX", "The MiniMax Open Platform documentation lists MiniMax-M2.1 as a served model at 204,800 context alongside its highspeed tier" ], "tradeoffs": [ "The governing license lives in a separate GitHub repository rather than alongside the weights, and its threshold-free attribution duty covers any commercial use", "Dated December 2025 repository history with no stated retirement posture, and development continues on a mutable main branch" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.82 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and in-repository deployment guides", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official MiniMax Open Platform text-generation documentation listing MiniMax-M2.1", "url": "https://platform.minimax.io/docs/guides/text-generation", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/minimax/MiniMax-M2.1.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-minimax-minimax-m2-5", "source_id": "minimax/MiniMax-M2.5", "name": "MiniMax-M2.5", "developer": "MiniMax", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "description": "MiniMax's 229-billion-parameter reasoning release for coding agents and complex workflows, published as downloadable weights under a bespoke model license for the weights and a modified MIT license for the repository code.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiniMax-M2.5", "description": "Prior MiniMax coding model for agent workflows, office edits, and automation", "family": "minimax", "release_date": "2026-02-12", "last_updated": "2026-02-12", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 204800, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5" } ] }, "licenses": [ "LicenseRef-MiniMax-M2.5-Model", "LicenseRef-MiniMax-M2.5-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The weight repository carries LICENSE-MODEL beside the shards while repository code is governed by a separate LICENSE-CODE file in the MiniMax-M2.5 GitHub repository; the card metadata points at a plain LICENSE path the weight repository does not carry. The model license grants use, reproduction, distribution, and modification of the model materials with a retained attribution notice, subject to a prohibited-uses policy and Singapore arbitration, while the code license reproduces the MIT grant with an unconditional duty to display 'MiniMax M2.5' on any commercial product or service. Both documents keep the release source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-MiniMax-M2.5-Model", "scope": "MiniMax-M2.5 model weights and accompanying materials beside the shards", "kind": "web_terms", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5/blob/f710177d938eff80b684d42c5aa84b382612f21f/LICENSE-MODEL", "verified_at": "2026-09-16" }, { "license_id": "LicenseRef-MiniMax-M2.5-Modified-MIT", "scope": "MiniMax-M2.5 repository code and derivative works in the MiniMax-M2.5 GitHub repository", "kind": "web_terms", "url": "https://github.com/MiniMax-AI/MiniMax-M2.5/blob/0fe00c843c16e7081a9631daeafc11288f5f871c/LICENSE-CODE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the MiniMaxAI/MiniMax-M2.5 weight release at Hugging Face commit f710177d938eff80b684d42c5aa84b382612f21f (last modified 2026-03-10), released under license terms dated 2026-02-13. The MiniMax-M2.1 predecessor, M2.7 and M3 successors, the M2.5-highspeed endpoint tier, the MiniMax Agent product, the platform API, and downstream quantizations remain separate boundaries.", "strengths": [ "126 Safetensors shards with LICENSE-MODEL beside the weights plus in-repository SGLang, vLLM, Transformers, and tool-calling guides", "The Open Platform lists MiniMax-M2.5 as a current served model at 204,800 context with a highspeed tier" ], "tradeoffs": [ "Two governing documents split across the weight repository and GitHub, with the card's license link not matching any file beside the weights", "No changelog or retirement posture accompanies the dated release" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 7.3 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and in-repository deployment guides", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official MiniMax Open Platform text-generation documentation listing MiniMax-M2.5", "url": "https://platform.minimax.io/docs/guides/text-generation", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/minimax/MiniMax-M2.5.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-minimax-minimax-m2-7", "source_id": "minimax/MiniMax-M2.7", "name": "MiniMax-M2.7", "developer": "MiniMax", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "description": "MiniMax's 229-billion-parameter reasoning release for agentic coding work, published as downloadable weights under a non-commercial license that requires prior written authorization for any commercial use.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiniMax-M2.7", "description": "Open MiniMax flagship for coding agents, office automation, and complex environments", "family": "minimax", "release_date": "2026-03-18", "last_updated": "2026-03-18", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 204800, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7" } ] }, "licenses": [ "LicenseRef-MiniMax-M2.7-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The full license text sits both beside the weights and in the MiniMax-M2.7 GitHub repository. It permits non-commercial use on MIT-style terms but prohibits any commercial use without separate prior written authorization from MiniMax, and additionally requires commercial deployments to display 'Built with MiniMax M2.7' — with no revenue or user threshold — so the release is source-available under terms stricter than M2's.", "license_evidence": [ { "license_id": "LicenseRef-MiniMax-M2.7-License", "scope": "MiniMax-M2.7 model weights, repository code, and derivative works, beside the weights and in the GitHub repository", "kind": "web_terms", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/d494266a4affc0d2995ba1fa35c8481cbd84294b/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the MiniMaxAI/MiniMax-M2.7 weight release at Hugging Face commit d494266a4affc0d2995ba1fa35c8481cbd84294b (last modified 2026-04-20). The MiniMax-M2.5 predecessor, the M3 generation, the M2.7-highspeed endpoint tier, the platform API, and downstream quantizations remain separate boundaries.", "strengths": [ "130 Safetensors shards with the full license text beside the weights and in-repository SGLang, vLLM, and Transformers deployment guides", "The Open Platform lists MiniMax-M2.7 as a served model at 204,800 context" ], "tradeoffs": [ "Any commercial use requires separate prior written authorization, the strictest commercial posture in the M2 line", "No changelog or retirement posture accompanies the release, and the platform lists a same-performance highspeed tier beside the line that carries no release identity of its own" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 6.88 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and in-repository deployment guides", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official MiniMax Open Platform text-generation documentation listing MiniMax-M2.7", "url": "https://platform.minimax.io/docs/guides/text-generation", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/minimax/MiniMax-M2.7.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-minimax-minimax-m3", "source_id": "minimax/MiniMax-M3", "name": "MiniMax-M3", "developer": "MiniMax", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3", "description": "MiniMax's 428-billion-parameter native multimodal release for agentic reasoning, coding, and long-context work, published as downloadable weights under a community license with a revenue-gated commercial authorization.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiniMax-M3", "description": "MiniMax multimodal model for long-context coding, perception, and agent planning", "family": "minimax", "release_date": "2026-06-01", "last_updated": "2026-06-01", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 512000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3" } ] }, "licenses": [ "LicenseRef-MiniMax-M3-Community" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The LICENSE file sits beside the weights and is the governing document; the M3 GitHub repository carries code and documentation but no license file of its own. The MiniMax Community License permits non-commercial use, requires commercial products and services to display 'Built with MiniMax M3', and requires separate prior written authorization above 20 million US dollars in yearly revenue — a one-time notice suffices below that line — so the release is source-available.", "license_evidence": [ { "license_id": "LicenseRef-MiniMax-M3-Community", "scope": "MiniMax-M3 model weights and accompanying materials beside the shards", "kind": "web_terms", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/f0e1c1e04d40177e4673a22097036854f536e9c0/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the MiniMaxAI/MiniMax-M3 weight release at Hugging Face commit f0e1c1e04d40177e4673a22097036854f536e9c0 (last modified 2026-07-23). The M2 generation lines, downstream quantizations and fine-tunes, the platform API, and independent hosting providers remain separate boundaries.", "strengths": [ "59 Safetensors shards with the community license beside the weights and in-repository evaluation results", "Native text, image, and video input with a 1M-token context window documented on the first-party platform" ], "tradeoffs": [ "Commercial use above 20 million US dollars in yearly revenue needs prior written authorization, and every commercial deployment carries the attribution duty", "No changelog or retirement posture accompanies the dated release" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.18 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and runtime guidance", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official MiniMax Open Platform text-generation documentation listing MiniMax-M3", "url": "https://platform.minimax.io/docs/guides/text-generation", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/minimax/MiniMax-M3.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-minimax-minimax-01", "source_id": null, "name": "MiniMax-Text-01", "developer": "MiniMax", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01", "description": "MiniMax's 456-billion-parameter hybrid lightning-attention mixture-of-experts language model with 45.9B active parameters and long-context inference, published as downloadable weights under the MiniMax Model License with a prohibited-uses policy and MIT-licensed code.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "MiniMax-Text-01", "description": "MiniMax's 456B-parameter hybrid-attention mixture-of-experts language model, 45.9B active parameters per token", "family": "minimax", "release_date": "2025-01-15", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 4000000, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "MiniMax Model License; code under MIT", "links": [ { "label": "GitHub", "url": "https://github.com/MiniMax-AI/MiniMax-01", "type": "docs" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01" } ] }, "licenses": [ "LicenseRef-MiniMax-Model-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The weights repository carries two files. LICENSE-MODEL, the MiniMax Model License dated for a model release date of 15 January 2025, grants use, reproduction, distribution, and modification of the Model Materials with a copy of the agreement and a Notice attribution, and incorporates a Prohibited Uses Policy whose items include any military purpose and fully automated decisions that adversely affect legal rights. That policy is a field-of-use restriction, so the release is source-available rather than open source. LICENSE-CODE is a separate MIT license for the repository's code, which is not the represented weights artifact and so is not listed as a record license; the weights remain under the restricted model terms. The model license is governed by Singapore law with SIAC arbitration, and the identical files appear in the GitHub repository.", "license_evidence": [ { "license_id": "LicenseRef-MiniMax-Model-License", "scope": "MiniMax-Text-01 model weights and accompanying materials, including the incorporated Prohibited Uses Policy", "kind": "web_terms", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/main/LICENSE-MODEL", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the MiniMaxAI/MiniMax-Text-01 weight release of January 15, 2025, in the original repository (last modified 2025-07-03). The MiniMax-VL-01 vision-language sibling, the transformers-native MiniMax-Text-01-hf conversion, the MiniMax-M1 reasoning release built on it, later MiniMax M-series models, downstream quantizations, and the MiniMax Open Platform are separate boundaries.", "strengths": [ "Both licence files sit beside the weights and in the GitHub repository, with the model terms dated by release date", "The model card documents a transformers quickstart with int8 quantization and a vLLM deployment guide" ], "tradeoffs": [ "The Prohibited Uses Policy bars military purposes and fully automated adverse decisions, so the licence restricts fields of use", "The current MiniMax platform documentation does not list this model, so no first-party hosted endpoint is documented today" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 2, "lifecycle_transparency": 5, "documentation_provenance": 8, "overall": 6.16 }, "evidence": [ { "kind": "web", "label": "Official MiniMax-Text-01 model card, checkpoint, and deployment guide", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax Model License and Prohibited Uses Policy in the weights repository", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/main/LICENSE-MODEL", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MIT code license in the weights repository", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/main/LICENSE-CODE", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax-01 GitHub repository with the same licence files", "url": "https://github.com/MiniMax-AI/MiniMax-01", "verified_at": "2026-10-06" }, { "kind": "web", "label": "MiniMax API docs release notes: MiniMax-Text-01 release entry of January 15, 2025", "url": "https://platform.minimax.io/docs/release-notes/models", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-mistral-codestral-2508", "source_id": null, "name": "Codestral 25.08", "developer": "Mistral AI", "url": "https://docs.mistral.ai/models/codestral-25-08", "description": "Mistral AI's July 2025 Codestral code-completion line, served as the codestral-2508 API model with fill-in-the-middle support, no downloadable weights, and no public weights license.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Codestral 25.08", "description": "Mistral AI code completion model for low-latency, high-frequency tasks such as fill-in-the-middle and code generation", "family": "codestral", "release_date": "2025-07-30", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 128000, "input": null, "output": null }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No public weights license exists for this API-only line: the docs page carries a Premier badge and no weights section, and no weight repository is published for codestral-2508. Use is governed by Mistral commercial and API terms, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of codestral-2508 with no public weights license", "kind": "web_terms", "url": "https://docs.mistral.ai/models/codestral-25-08", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the July 2025 Codestral 25.08 release served as the codestral-2508 API model (and the Codestral alias that currently resolves to it). The retired codestral-2501 API line, the Codestral-22B-v0.1 weight release, Codestral Embed, and the Mistral API platform itself are separate boundaries.", "strengths": [ "First-party docs pin the release with an API model ID, a GA status, and the fill-in-the-middle endpoint", "Dated release, structured outputs, function calling, and batch support are listed on one docs page" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; commercial terms govern entirely, and the docs page states no deprecation horizon" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 4, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 4.38 }, "evidence": [ { "kind": "web", "label": "Official Codestral 25.08 docs: release date, API ID, tier, and endpoints", "url": "https://docs.mistral.ai/models/codestral-25-08", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Mistral changelog entry for the Codestral 2508 release", "url": "https://docs.mistral.ai/getting-started/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Codestral 25.01 docs: predecessor retirement and replacement", "url": "https://docs.mistral.ai/models/codestral-25-01", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-mistral-codestral-22b-v0-1", "source_id": "mistral/codestral-22b-v0.1", "name": "Codestral-22B-v0.1", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1", "description": "Mistral AI's 22-billion-parameter coding release with fill-in-the-middle support, published as ungated weights under the non-production MNPL with a now-retired managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Codestral-22B-v0.1", "description": "Open Mistral code model for fill-in-the-middle and 80+ programming languages", "family": "codestral", "release_date": "2024-05-29", "last_updated": "2024-05-29", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Mistral AI Non-Production License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1" } ] }, "licenses": [ "LicenseRef-Mistral-Non-Production-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the file was deleted in May 2024, so the governing terms are the Mistral AI Non-Production License linked from the model card, news post, and docs page. The MNPL limits all use to testing, research, personal, or evaluation purposes in non-production environments and bars commercial supply, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Mistral-Non-Production-License", "scope": "Codestral-22B-v0.1 weights and derivatives, linked from the model card, news post, and docs page", "kind": "web_terms", "url": "https://mistral.ai/static/licenses/MNPL-0.1.md", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the mistralai/Codestral-22B-v0.1 weight release of May 2024 at Hugging Face commit 28b1c1a51dabe9d86ca8c41420ada1984632498f (last modified 2025-07-24). Its codestral-2405 API line retired on June 16, 2025. Later Codestral lines, other Mistral code models, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 22B coding checkpoint with tokenizer, params, and fill-in-the-middle support", "Release, retirement, and replacement all dated across first-party card, news, and docs" ], "tradeoffs": [ "Non-production-only grant bars any commercial use without a separate Mistral license", "The managed API is retired, leaving self-operation as the supported path" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 6.94 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Codestral docs: release date, retirement, and MNPL badge", "url": "https://docs.mistral.ai/models/codestral-24-05", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Codestral launch announcement (May 2024)", "url": "https://mistral.ai/news/codestral", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/codestral-22b-v0.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-devstral-2512", "source_id": "mistral/devstral-2512", "name": "Devstral 2", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512", "description": "Mistral AI's 123-billion-parameter dense Devstral 2 coding release, published as ungated weights under a Modified MIT license with a managed API now deprecated toward Medium 3.5.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Devstral 2", "description": "Mistral's coding-agent model for repository work, terminal tasks, and software fixes", "family": "devstral", "release_date": "2025-12-09", "last_updated": "2025-12-09", "knowledge_cutoff": "2025-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512" } ] }, "licenses": [ "LicenseRef-Mistral-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository LICENSE file is the Modified MIT text with a $20-million monthly-revenue gate on exercising any rights, matching the reviewed Medium 3.5 terms. Commercial users above that line need a separate Mistral commercial license or the hosted service, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Mistral-Modified-MIT", "scope": "Devstral-2-123B-Instruct-2512 model weights and derivatives beside the weights", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/1613bf01adb5e1c6fdc196b46e6b173eae75eb4a/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the mistralai/Devstral-2-123B-Instruct-2512 weight release of December 2025 at Hugging Face commit 1613bf01adb5e1c6fdc196b46e6b173eae75eb4a, together with the devstral-2512 API model, deprecated with migration directed at Mistral Medium 3.5. The Small 2 sibling line, Devstral Medium, other Mistral lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 123B dense checkpoint carrying its Modified MIT license file beside the weights", "First-party API with a 256K context window and dated deprecation plus replacement" ], "tradeoffs": [ "The revenue gate forces larger commercial users onto a separate license or the hosted service", "123B dense parameters need datacenter-class multi-GPU serving" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 4, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 6.94 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Devstral 2 docs: release date, API ID, and deprecation", "url": "https://docs.mistral.ai/models/devstral-2-25-12", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/devstral-2512.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-devstral-medium-2507", "source_id": "mistral/devstral-medium-2507", "name": "Devstral Medium", "developer": "Mistral AI", "url": "https://docs.mistral.ai/models/devstral-medium-1-0-25-07", "description": "Mistral AI's July 2025 Devstral Medium enterprise coding line, served as the devstral-medium-2507 API model with no public weights and no weights license.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Devstral Medium", "description": "Mistral coding agent model for repository tasks and software engineering workflows", "family": "devstral", "release_date": "2025-07-10", "last_updated": "2025-07-10", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No public weights license exists for this API-only line: no weight repository is published under any Mistral ID, and the docs page badges it Premier commercial. Use is governed by Mistral commercial and API terms, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of devstral-medium-2507 with no public weights license", "kind": "web_terms", "url": "https://docs.mistral.ai/models/devstral-medium-1-0-25-07", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the July 2025 Devstral Medium line served as the devstral-medium-2507 API model, deprecated with migration away from the Premier tier. No weight repository exists for this line. The Devstral 2 and Small 2 weight lines, other Mistral lines, and the Mistral API platform itself are separate boundaries.", "strengths": [ "First-party docs pin the July 2025 release with a deprecation date and Premier commercial terms", "Managed API plus Mistral Code and finetuning surfaces for enterprise deployment" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; commercial terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 2, "deployment_portability": 4, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.98 }, "evidence": [ { "kind": "web", "label": "Official Devstral Medium docs: release date, API ID, and deprecation", "url": "https://docs.mistral.ai/models/devstral-medium-1-0-25-07", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Devstral launch announcement (July 2025)", "url": "https://mistral.ai/news/devstral-2507/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/devstral-medium-2507.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-devstral-small-2507", "source_id": "mistral/devstral-small-2507", "name": "Devstral Small", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Devstral-Small-2507", "description": "A downloadable text-only coding-agent model derived from Mistral Small 3.1, published by Mistral AI with All Hands AI under an Apache-2.0 license stated in the model card and with documented local and vLLM serving paths.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Devstral Small", "description": "Mistral coding agent model for repository tasks and software engineering workflows", "family": "devstral", "release_date": "2025-07-10", "last_updated": "2025-07-10", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 128000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Devstral-Small-2507" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card states Apache 2.0 as the license for the Devstral-Small-2507 repository and checkpoints, both in its metadata and in prose. The repository carries no separate license file, so the pinned model card is the governing artifact; no use restrictions are attached.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official Devstral-Small-2507 model repository, consolidated and sharded Safetensors checkpoints, tokenizer, and parameters", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Devstral-Small-2507/blob/bd165ab26cebbcc2eea2c4ecbfc07f3ac42b3c39/README.md", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the mistralai/Devstral-Small-2507 release (Devstral Small 1.1, July 2025). The earlier Devstral-Small-2505, Devstral Medium, the Mistral Small 3.1 base, the deprecated devstral-small-2507 API model, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Mistral AI publishes ungated consolidated and Transformers-style Safetensors checkpoints, the tokenizer, the system prompt, and a first-party GGUF repository", "The model card documents vLLM, mistral-inference, Transformers, LM Studio, llama.cpp, and Ollama paths with pinned library versions and local agent-scaffold setup" ], "tradeoffs": [ "The license is carried in model-card metadata and prose rather than a repository license file", "The first-party API model named on the card is deprecated with a published retirement date, so served access now depends on self-operation or independent hosts" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 6, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.92 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and deployment guidance", "url": "https://huggingface.co/mistralai/Devstral-Small-2507", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Mistral AI model page with version, release, and deprecation dates", "url": "https://docs.mistral.ai/models/devstral-small-1-1-25-07", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/mistral/devstral-small-2507.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-mistral-devstral-small-2", "source_id": "mistral/devstral-small-2", "name": "Devstral Small 2", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512", "description": "An ungated 24B agentic coding model handling text and image input with a 262,144-token context, released under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Devstral Small 2", "description": "Compact multimodal coding model for repository exploration, file editing, and software agents", "family": "devstral", "release_date": "2025-12-09", "last_updated": "2025-12-09", "knowledge_cutoff": "2025-12", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card states the model is licensed under the Apache 2.0 License and carries an apache-2.0 frontmatter declaration, following the Mistral weight-release posture. The Mistral API is governed separately by the commercial terms of service.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/blob/55c5b41e98c2dbd21b0c8afffc540dcfc9eb5128/README.md", "verified_at": "2026-09-18", "scope": "Official mistralai/Devstral-Small-2-24B-Instruct-2512 weights, tokenizer, and bundled model code per the pinned model card" } ], "access_boundary": "This record represents the mistralai/Devstral-Small-2-24B-Instruct-2512 release (Devstral Small 2, December 2025). The Devstral 2 flagship, the Devstral-Small-2507 release, the retired labs-devstral-small-2512 API model, community quantizations and fine-tunes, and downstream products remain separate boundaries.", "strengths": [ "Ungated weights with consolidated and sharded Safetensors, tokenizer, and system prompt in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the Apache-2.0 grant rests on the card declaration", "The API serving line for this release is already deprecated or retired per the docs lifecycle table" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 6, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.92 }, "evidence": [ { "kind": "web", "label": "Official Devstral Small 2 weights repository and model card", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Devstral Small 2 model documentation", "url": "https://docs.mistral.ai/models/devstral-small-2-25-12", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Devstral Small 2 release announcement", "url": "https://mistral.ai/news/devstral-2-vibe-cli/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/devstral-small-2.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-magistral-small-2506", "source_id": "mistral/magistral-small-2506", "name": "Magistral Small", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Magistral-Small-2506", "description": "A downloadable text-only reasoning model derived from Mistral Small 3.1, published by Mistral AI with consolidated and sharded Safetensors checkpoints under an Apache-2.0 license stated in the model card.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Magistral Small", "description": "Open Mistral reasoning model for transparent step-by-step problem solving", "family": "magistral", "release_date": "2025-06-10", "last_updated": "2025-06-10", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Magistral-Small-2506" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card states Apache 2.0 as the license for the Magistral-Small-2506 repository and checkpoints, both in its metadata and in prose. The repository carries no separate license file, so the pinned model card is the governing artifact; no use restrictions are attached.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official Magistral-Small-2506 model repository, consolidated and sharded Safetensors checkpoints, tokenizer, and parameters", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Magistral-Small-2506/blob/ad2fce5b4113139e1877dcadfd826c9262ad8e8c/README.md", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the mistralai/Magistral-Small-2506 release (Magistral Small 1.0, June 2025). Magistral Medium, the Mistral Small 3.1 base, the later Magistral Small 2507 and 2509 releases, the retired magistral-small-2506 API model, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Mistral AI publishes ungated consolidated and Transformers-style Safetensors checkpoints, the tokenizer, the RL system prompt, and a first-party GGUF repository", "The model card documents a vLLM serving path with a pinned tokenizer-library version and names llama.cpp, LM Studio, Ollama, and Unsloth quantized paths" ], "tradeoffs": [ "The license is carried in model-card metadata and prose rather than a repository license file", "The first-party API model for this release is retired, so served access now depends on self-operation or independent hosts" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.46 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and deployment guidance", "url": "https://huggingface.co/mistralai/Magistral-Small-2506", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Mistral AI model page with version, release, and retirement dates", "url": "https://docs.mistral.ai/models/magistral-small-1-0-25-06", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/mistral/magistral-small-2506.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-mistral-magistral-small-2509", "source_id": "mistral/magistral-small-2509", "name": "Magistral Small 2509", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Magistral-Small-2509", "description": "An ungated 24B reasoning model handling text and image input with a 131,072-token context, released under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Magistral Small 1.2", "description": "Open multimodal reasoning model for transparent analysis of text and images", "family": "magistral", "release_date": "2025-09-18", "last_updated": "2025-09-18", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Magistral-Small-2509" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card states the model is licensed under the Apache 2.0 License and carries an apache-2.0 frontmatter declaration, following the Mistral weight-release posture. The Mistral API is governed separately by the commercial terms of service.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/README.md", "verified_at": "2026-09-18", "scope": "Official mistralai/Magistral-Small-2509 weights, tokenizer, and bundled model code per the pinned model card" } ], "access_boundary": "This record represents the mistralai/Magistral-Small-2509 release (Magistral Small 1.2, September 2025). The Magistral Small 2506 and 2507 releases, Magistral Medium, the retired magistral-small-2509 API model, community quantizations and fine-tunes, and downstream products remain separate boundaries.", "strengths": [ "Ungated weights with consolidated and sharded Safetensors, tokenizer, and system prompt in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the Apache-2.0 grant rests on the card declaration", "The API serving line for this release is already deprecated or retired per the docs lifecycle table" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.58 }, "evidence": [ { "kind": "web", "label": "Official Magistral Small 2509 weights repository and model card", "url": "https://huggingface.co/mistralai/Magistral-Small-2509", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Magistral Small 1.2 model documentation", "url": "https://docs.mistral.ai/models/magistral-small-1-2-25-09", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/magistral-small-2509.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-ministral-3-14b-instruct-2512", "source_id": "mistral/ministral-3-14b-instruct-2512", "name": "Ministral 3 14B Instruct", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512", "description": "An ungated 3 multimodal instruction model with a 262,144-token context, released under Apache-2.0 and served on the Mistral API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Ministral 3 14B", "description": "Open vision-language model for efficient local deployment, instruction following, and tool use", "family": "ministral", "release_date": "2025-12-02", "last_updated": "2025-12-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card states the model is licensed under the Apache 2.0 License and carries an apache-2.0 frontmatter declaration, following the Mistral weight-release posture. The Mistral API is governed separately by the commercial terms of service.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/README.md", "verified_at": "2026-09-18", "scope": "Official mistralai/Ministral-3-14B-Instruct-2512 weights, tokenizer, and bundled model code per the pinned model card" } ], "access_boundary": "This record represents the mistralai/Ministral-3-14B-Instruct-2512 release (December 2025). The Base and Reasoning siblings, official BF16 and GGUF checkpoints, the 3B and 8B lines, the retired 2410 line, the ministral-14b alias ID, the Studio API product, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated weights with consolidated and sharded Safetensors, tokenizer, and system prompt in the official repository", "The model card and launch announcement document the Apache-2.0 grant and wide hosting availability" ], "tradeoffs": [ "No LICENSE file ships with the weights; the Apache-2.0 grant rests on the card declaration", "No deprecation history exists yet for the line" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.94 }, "evidence": [ { "kind": "web", "label": "Official Ministral 3 14B Instruct weights repository and model card", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Ministral 3 14B model documentation", "url": "https://docs.mistral.ai/models/ministral-3-14b-25-12", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Ministral 3 launch announcement", "url": "https://mistral.ai/news/mistral-3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/ministral-3-14b-instruct-2512.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-ministral-3-3b-instruct-2512", "source_id": "mistral/ministral-3-3b-instruct-2512", "name": "Ministral 3 3B Instruct", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512", "description": "An ungated 3 multimodal instruction model with a 262,144-token context, released under Apache-2.0 and served on the Mistral API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Ministral 3 3B", "description": "Compact open vision-language model for edge deployment, instruction following, and tool use", "family": "ministral", "release_date": "2025-12-02", "last_updated": "2025-12-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card states the model is licensed under the Apache 2.0 License and carries an apache-2.0 frontmatter declaration, following the Mistral weight-release posture. The Mistral API is governed separately by the commercial terms of service.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/README.md", "verified_at": "2026-09-18", "scope": "Official mistralai/Ministral-3-3B-Instruct-2512 weights, tokenizer, and bundled model code per the pinned model card" } ], "access_boundary": "This record represents the mistralai/Ministral-3-3B-Instruct-2512 release (December 2025). The Base and Reasoning siblings, official BF16 and GGUF checkpoints, the 8B and 14B lines, the retired 2410 line, the Studio API product, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated weights with consolidated and sharded Safetensors, tokenizer, and system prompt in the official repository", "The model card and launch announcement document the Apache-2.0 grant and wide hosting availability" ], "tradeoffs": [ "No LICENSE file ships with the weights; the Apache-2.0 grant rests on the card declaration", "No deprecation history exists yet for the line" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.94 }, "evidence": [ { "kind": "web", "label": "Official Ministral 3 3B Instruct weights repository and model card", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Ministral 3 3B model documentation", "url": "https://docs.mistral.ai/models/ministral-3-3b-25-12", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Ministral 3 launch announcement", "url": "https://mistral.ai/news/mistral-3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/ministral-3-3b-instruct-2512.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-ministral-3-8b-instruct-2512", "source_id": "mistral/ministral-3-8b-instruct-2512", "name": "Ministral 3 8B Instruct", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512", "description": "An ungated 3 multimodal instruction model with a 262,144-token context, released under Apache-2.0 and served on the Mistral API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Ministral 3 8B", "description": "Compact open vision-language model for edge deployment, instruction following, and tool use", "family": "ministral", "release_date": "2025-12-02", "last_updated": "2025-12-02", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card states the model is licensed under the Apache 2.0 License and carries an apache-2.0 frontmatter declaration, following the Mistral weight-release posture. The Mistral API is governed separately by the commercial terms of service.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/README.md", "verified_at": "2026-09-18", "scope": "Official mistralai/Ministral-3-8B-Instruct-2512 weights, tokenizer, and bundled model code per the pinned model card" } ], "access_boundary": "This record represents the mistralai/Ministral-3-8B-Instruct-2512 release (December 2025). The Base and Reasoning siblings, official BF16 and GGUF checkpoints, the 3B and 14B lines, the retired 2410 line, the Studio API product, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated weights with consolidated and sharded Safetensors, tokenizer, and system prompt in the official repository", "The model card and launch announcement document the Apache-2.0 grant and wide hosting availability" ], "tradeoffs": [ "No LICENSE file ships with the weights; the Apache-2.0 grant rests on the card declaration", "No deprecation history exists yet for the line" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.94 }, "evidence": [ { "kind": "web", "label": "Official Ministral 3 8B Instruct weights repository and model card", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Ministral 3 8B model documentation", "url": "https://docs.mistral.ai/models/ministral-3-8b-25-12", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Ministral 3 launch announcement", "url": "https://mistral.ai/news/mistral-3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/ministral-3-8b-instruct-2512.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-ministral-8b-instruct-2410", "source_id": "mistral/ministral-8b-instruct-2410", "name": "Ministral 8B Instruct", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410", "description": "Mistral AI's 8-billion-parameter Ministral edge release, published as gated weights under the Mistral Research License with a now-retired managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ministral 8B Instruct", "description": "Efficient open Mistral edge model for on-device chat and function calling", "family": "ministral", "release_date": "2024-10-16", "last_updated": "2024-10-16", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Mistral Research License", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410" } ] }, "licenses": [ "LicenseRef-Mistral-Research-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the gated card links the Mistral Research License as the governing terms. The MRL grants use, copying, modification, and distribution but limits all use of the model, derivatives, and outputs to non-profit, non-commercial research purposes, with commercial use requiring a separate license from Mistral AI, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Mistral-Research-License", "scope": "Ministral-8B-Instruct-2410 weights and derivatives, linked from the gated model card", "kind": "web_terms", "url": "https://mistral.ai/licenses/MRL-0.1.md", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the October 2024 Ministral-8B-Instruct-2410 weight release at Hugging Face commit 2f494a194c5b980dfb9772cb92d26cbb671fce5a (last modified 2025-07-31). Its ministral-8b-2410 API line retired on December 31, 2025 with migration directed at Ministral 3 8B. The Ministral 3B sibling, later Ministral lines, and community quantizations remain separate boundaries.", "strengths": [ "Gated 8B edge checkpoint under explicit research-only terms, runnable on 24 GB single-GPU setups", "Release, retirement, and replacement all dated across first-party card, news, and docs" ], "tradeoffs": [ "Research-only grant forces any commercial use onto a separate Mistral commercial license", "The managed API is retired, leaving gated self-operation as the supported path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 7, "deployment_portability": 5, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 5.94 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Ministral launch announcement (October 2024)", "url": "https://mistral.ai/news/ministraux/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/ministral-8b-instruct-2410.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-mistral-mistral-large-2407", "source_id": null, "name": "Mistral Large 2", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2407", "description": "Mistral AI's 123-billion-parameter dense Large 2 release of July 2024 for research use, published as gated weights under the Mistral Research License with a now-retired managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Mistral Large 2", "description": "123B dense multilingual model with native function calling, a 128k context window, and research-only open weights", "family": "mistral-large", "release_date": "2024-07-24", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 128000, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2407" } ] }, "licenses": [ "LicenseRef-Mistral-Research-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "No LICENSE file ships in the gated weight repository; the model card embeds and links the Mistral Research License as the governing terms. The MRL grants use, copying, modification, and distribution but limits all use of the model, derivatives, and outputs to non-profit, non-commercial research purposes, with commercial use requiring a separate license from Mistral AI, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Mistral-Research-License", "scope": "Mistral-Large-Instruct-2407 weights and derivatives, embedded in the gated model card", "kind": "web_terms", "url": "https://mistral.ai/licenses/MRL-0.1.md", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the mistralai/Mistral-Large-Instruct-2407 weight release of July 2024 at Hugging Face commit a286006d554cb37a61d13c7ae61bc90cc1d372fc (last modified 2025-07-28). Its mistral-large-2407 API line retired on March 30, 2025 with migration directed at Mistral Large 3. The Large 2.1 and Large 3 records, the retired Large 1.0 API line, and community quantizations remain separate boundaries.", "strengths": [ "Gated 123B dense checkpoint under explicit research-only terms with a 128K context window", "Release, retirement, and replacement all dated on first-party docs" ], "tradeoffs": [ "Research-only grant forces any commercial use onto a separate Mistral commercial license", "The managed API is retired, leaving self-operation of a 123B model as the supported path" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 6.76 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2407", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Mistral Large 2.0 docs: release date, API ID, MRL badge, and retirement", "url": "https://docs.mistral.ai/models/mistral-large-2-0-24-07", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Mistral AI Research License text", "url": "https://mistral.ai/licenses/MRL-0.1.md", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-mistral-mistral-large-2411", "source_id": "mistral/mistral-large-2411", "name": "Mistral Large 2.1", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411", "description": "Mistral AI's 123-billion-parameter dense Large release for research use, published as gated weights under the Mistral Research License with managed API and cloud hosting.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Mistral Large 2.1", "description": "Flagship Mistral model for advanced reasoning, coding, and multilingual work", "family": "mistral-large", "release_date": "2024-11-18", "last_updated": "2024-11-18", "knowledge_cutoff": "2024-11", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411" } ] }, "licenses": [ "LicenseRef-Mistral-Research-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the gated card links the Mistral Research License as the governing terms. The MRL grants use, copying, modification, and distribution but limits all use of the model, derivatives, and outputs to non-profit, non-commercial research purposes, with commercial use requiring a separate license from Mistral AI, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Mistral-Research-License", "scope": "Mistral-Large-Instruct-2411 weights and derivatives, linked from the gated model card", "kind": "web_terms", "url": "https://mistral.ai/licenses/MRL-0.1.md", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the mistralai/Mistral-Large-Instruct-2411 weight release of November 2024 at Hugging Face commit ba78820945ae22361b0274cf0ae6d696c967c1a4 (last modified 2025-07-28), together with the mistral-large-2411 API model and its Vertex AI and GitHub Models hosting. Later Large generations, the Pixtral releases of the same announcement, and community quantizations remain separate boundaries.", "strengths": [ "Gated 123B dense checkpoint under explicit research-only terms with a 128K context window", "First-party API plus documented Vertex AI and GitHub Models hosting" ], "tradeoffs": [ "Research-only grant forces any commercial use onto a separate Mistral commercial license", "The line is deprecated with migration directed at Mistral Medium 3.5" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.24 }, "evidence": [ { "kind": "web", "label": "Official model card and gated weight release", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Mistral Large 2.1 docs: release date, API ID, and deprecation", "url": "https://docs.mistral.ai/models/mistral-large-2-1-24-11", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Google Cloud announcement of Mistral Large 24.11 on Vertex AI", "url": "https://cloud.google.com/blog/products/ai-machine-learning/announcing-new-mistral-large-model-on-vertex-ai", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/mistral-large-2411.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-mistral-mistral-large-2512", "source_id": "mistral/mistral-large-2512", "name": "Mistral Large 3", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512", "description": "Mistral AI's 675-billion-parameter granular Mixture-of-Experts multimodal flagship with 41B active parameters, published as Apache-2.0 weights with managed API and cloud hosting.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Mistral Large 3", "description": "Mistral's largest general model for enterprise agents, coding, and multilingual reasoning", "family": "mistral-large", "release_date": "2025-12-02", "last_updated": "2025-12-02", "knowledge_cutoff": "2024-11", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the model card frontmatter, badge, and prose declare the Apache 2.0 license, and the GA docs page badges the release Apache 2.0, with no use restrictions attached — the same card-only pattern as the reviewed Small 4 record.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Mistral Large 3 weights as declared on the official model card and GA docs page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the December 2025 Mistral Large 3 release at Hugging Face commit 383ffea2c7d60dfd44ca960e8e691709d4fdb9cd (last modified 2026-07-15), covering the FP8, NVFP4, and BF16 checkpoints, together with the mistral-large-2512 API model and its Azure and Bedrock hosting. Earlier Large generations, the base weights, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 675B MoE checkpoints in three precisions under a declared Apache-2.0 grant", "GA developer API plus documented Azure and Bedrock hosting with a 256K context window" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card and docs declarations", "675B total parameters need single-node 8-GPU-class serving even in FP8" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.6 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Mistral Large 3 docs: GA status, Apache-2.0 badge, and API ID", "url": "https://docs.mistral.ai/models/mistral-large-3-25-12", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Microsoft announcement of Mistral Large 3 on Azure", "url": "https://azure.microsoft.com/en-us/blog/introducing-mistral-large-3-in-microsoft-foundry-open-capable-and-ready-for-production-workloads", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/mistral-large-2512.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-mistral-mistral-large-4", "source_id": "mistral/mistral-large-4", "name": "Mistral Large 4", "developer": "Mistral AI", "url": "https://docs.mistral.ai/models/mistral-large-4-0", "description": "Mistral AI's roughly one-trillion-parameter natively multimodal Mixture-of-Experts flagship, available today as a public-preview API model with open weights announced for the end of October 2026.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Mistral Large 4", "description": "Mistral's 1T-parameter multimodal MoE for reasoning, coding, and agentic work", "family": "mistral-large", "release_date": "2026-10-06", "last_updated": "2026-10-06", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 262144 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "At review time no weights are published and no weights license exists: the announcement says weights will be released by the end of October 2026, and the docs page shows no license badge. The preview API is governed by Mistral's commercial terms, so the release is classified proprietary today. Re-review when the weights and their license text ship.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "Public-preview API access to Mistral Large 4; no weights or weights license published as of the review date", "kind": "web_terms", "url": "https://mistral.ai/news/mistral-large-4", "verified_at": "2026-10-07" } ], "access_boundary": "This record represents Mistral Large 4 (announced 2026-10-06) as the public-preview mistral-large-4 API model (alias mistral-large-4-0). The weights promised for the end of October 2026, the reduced-moderation variant offered to vetted red-teaming partners, earlier Large generations, and downstream quantizations remain separate boundaries. The announcement states 1 trillion parameters with 49B active; the docs page states 1.05T total with 52B active and a 1.6B vision encoder.", "strengths": [ "First-party announcement and docs page published on launch day with a 1M-token context window and documented function calling, structured outputs, batching, and agents", "Mistral commits to publishing the weights by the end of October 2026 and to serving from European infrastructure it operates itself" ], "tradeoffs": [ "Public preview only: no weights, license text, or third-party hosting exist yet, so artifact availability and portability rest on a promise", "The announcement and docs page disagree on active-parameter count (49B versus 52B)" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.0 }, "evidence": [ { "kind": "web", "label": "Official Mistral Large 4 announcement: preview API and end-of-month weights commitment", "url": "https://mistral.ai/news/mistral-large-4", "verified_at": "2026-10-07" }, { "kind": "web", "label": "Official Mistral Large 4 docs page: public-preview status, context window, and features", "url": "https://docs.mistral.ai/models/mistral-large-4-0", "verified_at": "2026-10-07" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/195b940bc687ae56cf2f2e68d42bc830d9cf653f/models/mistral/mistral-large-4.toml", "verified_at": "2026-10-07" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://mistral.ai/news/mistral-large-4", "verified_at": "2026-10-07", "research_confidence": "medium" } ], "metadata_verified_at": "2026-10-07", "verified_at": "2026-10-07" }, { "id": "model-mistral-mistral-medium-2505", "source_id": "mistral/mistral-medium-2505", "name": "Mistral Medium 3", "developer": "Mistral AI", "url": "https://docs.mistral.ai/models/mistral-medium-3-25-05", "description": "Mistral AI's May 2025 frontier multimodal line, served as the mistral-medium-2505 API model with no downloadable weights and no public weights license.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Mistral Medium 3", "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows", "family": "mistral-medium", "release_date": "2025-05-07", "last_updated": "2025-05-07", "knowledge_cutoff": "2025-05", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No public weights license exists for this API-only line: no weight repository carries a license file, and the docs page shows no license badge. Use is governed by Mistral commercial and API terms, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of mistral-medium-2505 with no public weights license", "kind": "web_terms", "url": "https://docs.mistral.ai/models/mistral-medium-3-25-05", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the May 2025 Mistral Medium line served as the mistral-medium-2505 API model, deprecated with migration directed at Mistral Medium 3.5. No weight repository exists under any Mistral ID for this line. The Medium 3.5 successor, other Mistral lines, and the Mistral API platform itself are separate boundaries.", "strengths": [ "First-party docs pin the May 2025 release with a deprecation date and named replacement", "Managed API serving with multimodal input at 128K context" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; commercial terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 2, "deployment_portability": 5, "serving_reach": 6, "lifecycle_transparency": 8, "documentation_provenance": 7, "overall": 5.02 }, "evidence": [ { "kind": "web", "label": "Official Mistral Medium docs: release date, API ID, and deprecation", "url": "https://docs.mistral.ai/models/mistral-medium-3-25-05", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/mistral-medium-2505.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-mistral-mistral-medium-3-1", "source_id": null, "name": "Mistral Medium 3.1", "developer": "Mistral AI", "url": "https://docs.mistral.ai/models/mistral-medium-3-1-25-08", "description": "Mistral AI's August 2025 frontier multimodal Medium update, served as the mistral-medium-2508 API model with no downloadable weights and no public weights license.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Mistral Medium 3.1", "description": "Frontier-class multimodal model released August 2025 with improved tone and performance", "family": "mistral-medium", "release_date": "2025-08-12", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 128000, "input": null, "output": null }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No public weights license exists for this API-only line: the docs page carries a Premier badge and no weights section, and no weight repository is published for Mistral Medium 3.1. Use is governed by Mistral commercial and API terms, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of mistral-medium-2508 with no public weights license", "kind": "web_terms", "url": "https://docs.mistral.ai/models/mistral-medium-3-1-25-08", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the August 2025 Mistral Medium 3.1 release served as the mistral-medium-2508 API model, deprecated on May 22, 2026 with migration directed at Mistral Medium 3.5. The Medium 3 and Medium 3.5 records, other Mistral lines, and the Mistral API platform itself are separate boundaries.", "strengths": [ "First-party docs pin the August 2025 release with a deprecation date and a named replacement", "Managed API with multimodal input at 128K context, plus agents and batch support" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; commercial terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 4, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 7, "overall": 4.5 }, "evidence": [ { "kind": "web", "label": "Official Mistral Medium 3.1 docs: release date, API ID, tier, and deprecation", "url": "https://docs.mistral.ai/models/mistral-medium-3-1-25-08", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Mistral changelog entry for the Medium 3.1 release", "url": "https://docs.mistral.ai/getting-started/changelog", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-mistral-mistral-medium-2604", "source_id": "mistral/mistral-medium-2604", "name": "Mistral Medium 3.5", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B", "description": "A downloadable image-and-text flagship model published by Mistral AI with consolidated and sharded Safetensors checkpoints under a repository Modified MIT License carrying a monthly-revenue restriction, and served through Mistral's developer API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Mistral Medium 3.5", "description": "Balanced Mistral model for enterprise assistants, multilingual work, and tools", "family": "mistral-medium", "release_date": "2026-04-29", "last_updated": "2026-04-29", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" } ] }, "licenses": [ "LicenseRef-Mistral-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository carries its own license file stating a Modified MIT License with a 2026 Mistral AI attribution notice. The MIT grant is conditioned on the licensee's global consolidated monthly revenue not exceeding 20 million US dollars for the preceding month, with a commercial-license contact offered as the alternative. That revenue restriction is not part of the OSI-approved MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Mistral-Modified-MIT", "scope": "Official Mistral-Medium-3.5-128B model repository, consolidated and sharded Safetensors checkpoints, tokenizer, parameters, and derivatives", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/blob/22b2b868a15677cfa6061277ed2f653d1349a9ab/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the mistralai/Mistral-Medium-3.5-128B release (Mistral Medium 3.5, April 2026; dense 128B, 256k context). The EAGLE speculative-decoding sibling repository, the Magistral and Devstral lines, the mistral-medium-3-5 API product and Le Chat deployments, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Mistral AI publishes ungated consolidated and Transformers-style Safetensors checkpoints with tokenizer, system prompt, and an in-repository license file", "The model card documents vLLM, SGLang, Transformers, llama.cpp, and Ollama paths, and Mistral serves the release as a GA priced API model that also ships as an NVIDIA NIM" ], "tradeoffs": [ "The Modified MIT License bars exercise of rights above a 20M US dollar monthly-revenue threshold, so the release is source-available rather than open source", "The 128B dense footprint needs multi-GPU tensor parallelism per the serving guide" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.42 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and deployment guidance", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Mistral AI model page with release date and license badge", "url": "https://docs.mistral.ai/models/mistral-medium-3-5-26-04", "verified_at": "2026-09-11" }, { "kind": "web", "label": "NVIDIA NGC catalog entry for the NIM container of the exact release", "url": "https://catalog.ngc.nvidia.com/orgs/nim/mistralai/containers/mistral-medium-3.5-128b/-", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/mistral/mistral-medium-2604.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-mistral-mistral-nemo", "source_id": "mistral/mistral-nemo", "name": "Mistral Nemo", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407", "description": "Mistral AI and NVIDIA's 12-billion-parameter multilingual release, published as ungated Apache-2.0 weights with managed API and NIM hosting, now deprecated toward Ministral 3 8B.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Mistral Nemo", "description": "Efficient Mistral-NVIDIA open model for multilingual chat and local deployment", "family": "mistral-nemo", "release_date": "2024-07-01", "last_updated": "2024-07-01", "knowledge_cutoff": "2024-07", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 128000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repositories; the launch announcement, model card, docs badge, and NVIDIA NIM page concordantly declare the Apache 2.0 license with no use restrictions attached — the same card-only pattern as the reviewed Small 4 record.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Mistral Nemo weights as declared on the official model card, launch announcement, docs badge, and NIM page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the July 2024 Mistral Nemo 12B (24.07) release across its base and instruct weight repositories at commits a4477a2f977929a969745b69bbd62e03043551a5 and 04d8a90549d23fc6bd7f642064003592df51e9b3 (both last modified 2025-07-28), together with the open-mistral-nemo-2407 API model and its NIM container. The FP8 quantization, NVIDIA's NeMo-format repackaging, and later Ministral lines remain separate boundaries.", "strengths": [ "Ungated 12B base and instruct checkpoints under a four-surface declared Apache-2.0 grant", "First-party API plus exact-model NVIDIA NIM hosting with a 128K context window" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on announcement, card, docs, and host declarations", "The managed API is deprecated and retired with migration directed at Ministral 3 8B" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.4 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Mistral Nemo launch announcement (July 2024)", "url": "https://mistral.ai/news/mistral-nemo/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Mistral Nemo docs: API ID, deprecation, and replacement", "url": "https://docs.mistral.ai/models/mistral-nemo-12b-24-07", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA NIM page for the exact mistral-nemo-12b-instruct model", "url": "https://build.nvidia.com/nv-mistralai/mistral-nemo-12b-instruct", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/mistral-nemo.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-mistral-small-24b-instruct-2501", "source_id": null, "name": "Mistral Small 3", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501", "description": "Mistral AI's 24-billion-parameter Small 3 instruct release of January 2025, published as Apache-2.0 weights with a now-retired managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Mistral Small 3", "description": "24B instruction-tuned model with native function calling and a 32k context window under Apache 2.0", "family": "mistral-small", "release_date": "2025-01-30", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 32000, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the model card frontmatter and prose both state the Apache 2.0 license, and the docs page badges the release Apache 2.0, with no use restrictions attached - the same card-only pattern as the reviewed Small 3.1 and Small 4 records.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Mistral Small 3 weights as declared on the official model card and docs page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the January 2025 Mistral Small 3 instruct release at Hugging Face commit 9527884be6e5616bdd54de542f9ae13384489724 (last modified 2025-07-28). Its mistral-small-2501 API line retired on November 30, 2025 with migration directed at Mistral Small 4. The Small 3 base weights, the Small 3.1 and Small 3.2 successors, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 24B checkpoint under a declared Apache-2.0 grant with no use restrictions", "Release, retirement, and replacement all dated on first-party docs" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card and docs declarations", "The managed API is retired, leaving self-operation as the supported path" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.58 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Mistral Small 3.0 docs: release date, Apache badge, and retirement", "url": "https://docs.mistral.ai/models/mistral-small-3-0-25-01", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Apache License 2.0 standard text", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-mistral-mistral-small-3-1-24b-instruct-2503", "source_id": "mistral/mistral-small-3-1-24b-instruct-2503", "name": "Mistral Small 3.1 24B", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "description": "Mistral AI's 24-billion-parameter Small 3.1 multimodal release with vision understanding, published as Apache-2.0 weights with a now-retired managed API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Mistral Small 3.1 24B", "description": "Efficient multimodal model for instruction following, coding, reasoning, and function calling", "family": "mistral-small", "release_date": "2025-03-17", "last_updated": "2025-03-17", "knowledge_cutoff": "2024-06", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the model card frontmatter and prose both state the Apache 2.0 license, and the docs page badges the release Apache 2.0, with no use restrictions attached — the same card-only pattern as the reviewed Small 4 record.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Mistral Small 3.1 weights as declared on the official model card and docs page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the March 2025 Mistral Small 3.1 release at Hugging Face commit 68faf511d618ef198fef186659617cfd2eb8e33a (last modified 2025-12-22). Its mistral-small-2503 API line retired on November 30, 2025 with migration directed at Mistral Small 4. The Small 3.2 successor, other Mistral lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 24B multimodal checkpoint under a declared Apache-2.0 grant with vision examples", "Release, retirement, and replacement all dated on first-party docs" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card and docs declarations", "The managed API is retired, leaving self-operation as the supported path" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.58 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Mistral Small 3.1 docs: release date, retirement, and replacement", "url": "https://docs.mistral.ai/models/mistral-small-3-1-25-03", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/mistral-small-3-1-24b-instruct-2503.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-mistral-mistral-small-2506", "source_id": "mistral/mistral-small-2506", "name": "Mistral Small 3.2", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506", "description": "Mistral AI's 24-billion-parameter Small 3.2 multimodal release, published as Apache-2.0 weights with a managed API now deprecated toward Small 4.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Mistral Small 3.2", "description": "Efficient Mistral model for fast chat, extraction, and production assistants", "family": "mistral-small", "release_date": "2025-06-20", "last_updated": "2025-06-20", "knowledge_cutoff": "2025-03", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the model card frontmatter declares Apache-2.0 and its prose inherits the Small 3.1 card's Apache-2.0 statement, and the docs page badges the release Apache 2.0, with no use restrictions attached — the same card-only pattern as the reviewed Small 4 record.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Mistral Small 3.2 weights as declared on the official model card and docs page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the June 2025 Mistral Small 3.2 release at Hugging Face commit 95a6d26c4bfb886c58daf9d3f7332c857cb27b43 (last modified 2025-12-22), together with the mistral-small-2506 API model, deprecated with migration directed at Mistral Small 4. The Small 3.1 predecessor, Small 4, other Mistral lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 24B multimodal checkpoint under a declared Apache-2.0 grant with tokenizer and system prompt", "First-party API with documented vLLM and Transformers serving paths" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card and docs declarations", "The API line is deprecated with migration directed at Mistral Small 4" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.72 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Mistral Small 3.2 docs: release date, API ID, and deprecation", "url": "https://docs.mistral.ai/models/mistral-small-3-2-25-06", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/mistral-small-2506.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-mistral-mistral-small-2603", "source_id": "mistral/mistral-small-2603", "name": "Mistral Small 4", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603", "description": "A downloadable image-and-text hybrid instruction-and-reasoning model published by Mistral AI with consolidated and sharded Safetensors checkpoints under an Apache-2.0 license stated in the model card, and served through Mistral's developer API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Mistral Small 4", "description": "Fast Mistral production model for chat, extraction, and cost-sensitive agents", "family": "mistral-small", "release_date": "2026-03-16", "last_updated": "2026-03-16", "knowledge_cutoff": "2025-06", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 256000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model card states Apache 2.0 as the license for the Mistral-Small-4-119B-2603 repository and checkpoints, both in its metadata and in prose. The repository carries no separate license file, so the pinned model card is the governing artifact; no use restrictions are attached. Mistral's first-party docs page independently badges the release Apache 2.0.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official Mistral-Small-4-119B-2603 model repository, consolidated and sharded Safetensors checkpoints, tokenizer, and parameters", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/blob/a11f36bebf709121056b1dbcc943d1c6afbe494d/README.md", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the mistralai/Mistral-Small-4-119B-2603 release (Mistral Small 4, March 16, 2026; 119B MoE, 256k context). Mistral Small 3.2-24B-Instruct-2506, the Magistral line, the eagle speculative-decoding head and NVFP4 sibling repositories, Mistral Medium 3.5, the mistral-small-2603 API product, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Mistral AI publishes ungated consolidated and Transformers-style Safetensors checkpoints with tokenizer, system prompt, and efficiency sibling repositories", "The model card documents vLLM, Transformers, SGLang, llama.cpp, and LM Studio paths, and Mistral serves the release as a GA priced API model" ], "tradeoffs": [ "The license is carried in model-card metadata and prose rather than a repository license file", "The 119B MoE footprint needs multi-GPU serving, so local use depends on the documented runtimes or hosted paths" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.86 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and deployment guidance", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Mistral AI model page with release date and Apache 2.0 badge", "url": "https://docs.mistral.ai/models/mistral-small-4-0-26-03", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Mistral AI launch announcement", "url": "https://mistral.ai/news/mistral-small-4/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "NVIDIA NIM reference documentation for the exact release", "url": "https://docs.api.nvidia.com/nim/reference/mistralai-mistral-small-4-119b-2603", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/mistral/mistral-small-2603.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-mistral-mixtral-8x22b-instruct", "source_id": null, "name": "Mixtral 8x22B Instruct", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1", "description": "Mistral AI's sparse mixture-of-experts Mixtral 8x22B instruct release of April 2024, published as Apache-2.0 weights with a now-retired managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Mixtral 8x22B Instruct", "description": "Open sparse mixture-of-experts model with 141B total and 39B active parameters, instruct fine-tuned with function-calling tokens", "family": "mixtral", "release_date": "2024-04-17", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 64000, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the model card frontmatter declares Apache 2.0 and the docs page badges the instruct weights Apache 2.0, with no use restrictions attached - the same card-only pattern as the reviewed Small 3.1 record.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Mixtral-8x22B-Instruct-v0.1 weights as declared on the official model card and docs page, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the mistralai/Mixtral-8x22B-Instruct-v0.1 instruct weight release at Hugging Face commit cc88a6cc19fbd17d9f1c0ee0b0d70a748dce698d (last modified 2025-07-24). Its open-mixtral-8x22b API line retired on March 30, 2025 with migration directed at Mistral Small 4. The Mixtral-8x22B-v0.1 base weights, Mixtral 8x7B, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 141B-total, 39B-active checkpoint under a declared Apache-2.0 grant with no use restrictions", "Release, retirement, and replacement all dated on first-party docs" ], "tradeoffs": [ "No LICENSE file ships beside the weights, and the card has no release-date or version statement, so the grant rests on card and docs declarations", "The managed API is retired, and the very large checkpoint narrows practical self-operation" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 7, "overall": 7.24 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Mixtral 8x22B docs: release date, Apache badge, and retirement", "url": "https://docs.mistral.ai/models/mixtral-8x22b-0-1-0-3", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Apache License 2.0 standard text", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-mistral-pixtral-12b", "source_id": "mistral/pixtral-12b", "name": "Pixtral 12B", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409", "description": "A downloadable image-and-text instruction model published by Mistral AI with a Safetensors checkpoint and an Apache-2.0 distribution boundary.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Pixtral 12B", "description": "Mistral vision-language model for image understanding and multimodal chat", "family": "pixtral", "release_date": "2024-09-01", "last_updated": "2024-09-01", "knowledge_cutoff": "2024-09", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 128000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official Mistral AI model card identifies Apache-2.0 as the license for the published Pixtral 12B repository and checkpoint. The repository exposes the license identifier in its pinned model card rather than a separate license file.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official Pixtral-12B-2409 model repository, consolidated Safetensors checkpoint, tokenizer, and parameters", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409/blob/a5ca2b7a9421a35ea24411c46e970758b8d3c6a7/README.md", "verified_at": "2026-09-04" } ], "access_boundary": "This record represents the Pixtral-12B-2409 instruction release. The base model, hosted Mistral services, downstream quantizations and fine-tunes, and later Pixtral releases remain separate boundaries.", "strengths": [ "Mistral AI publishes a directly downloadable consolidated checkpoint and tokenizer artifacts", "The official model card documents a self-operated vLLM path and names the distribution license" ], "tradeoffs": [ "The license statement is carried in model-card metadata instead of a repository license file", "The model card recommends a particular serving stack, so portability beyond documented paths needs independent runtime review" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.6 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and vLLM deployment guidance", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/ef71a4865a5821d8cd24f171a0100156a02e808e/models/mistral/pixtral-12b.toml", "verified_at": "2026-09-04" } ], "metadata_verified_at": "2026-09-04", "verified_at": "2026-09-04" }, { "id": "model-mistral-voxtral-mini-3b-2507", "source_id": "mistral/voxtral-mini-3b-2507", "name": "Voxtral Mini 3B 2507", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507", "description": "An ungated 4.7B audio-and-text input model producing text output with a 32,768-token context, released under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Voxtral Mini 3B 2507", "description": "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use", "family": "voxtral", "release_date": "2025-07-15", "last_updated": "2025-07-15", "knowledge_cutoff": null, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries an apache-2.0 frontmatter declaration corroborated by the Hugging Face tag, the launch announcement, and the docs page badge. The Mistral API is governed separately by the commercial terms of service.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507/blob/3060fe34b35ba5d44202ce9ff3c097642914f8f3/README.md", "verified_at": "2026-09-18", "scope": "Official mistralai/Voxtral-Mini-3B-2507 weights, tokenizer, and bundled model code per the pinned model card" } ], "access_boundary": "This record represents the mistralai/Voxtral-Mini-3B-2507 release (July 2025). The Voxtral Small sibling, the transcribe-optimized API product, the retired voxtral-mini-2507 API model, community quantizations and fine-tunes, and downstream products remain separate boundaries.", "strengths": [ "Ungated audio-input weights with consolidated and sharded Safetensors and tokenizer in the official repository", "The model card documents the audio length limits and Transformers and vLLM deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the Apache-2.0 grant rests on the card declaration", "The 32K token context bounds the audio input to the documented duration limits" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 6, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.92 }, "evidence": [ { "kind": "web", "label": "Official Voxtral Mini 2507 weights repository and model card", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Voxtral Mini model documentation", "url": "https://docs.mistral.ai/models/voxtral-mini-25-07", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Voxtral launch announcement", "url": "https://mistral.ai/news/voxtral", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/voxtral-mini-3b-2507.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mistral-voxtral-small-24b-2507", "source_id": "mistral/voxtral-small-24b-2507", "name": "Voxtral Small 24B 2507", "developer": "Mistral AI", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507", "description": "An ungated 24.3B audio-and-text input model producing text output with a 32,768-token context, released under Apache-2.0 and served on the Mistral API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Voxtral Small 24B 2507", "description": "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use", "family": "voxtral", "release_date": "2025-07-15", "last_updated": "2025-07-15", "knowledge_cutoff": null, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries an apache-2.0 frontmatter declaration corroborated by the Hugging Face tag, the launch announcement, and the docs page badge. The Mistral API is governed separately by the commercial terms of service.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507/blob/da5b42409f279fdd92febee0511a6c32828569c1/README.md", "verified_at": "2026-09-18", "scope": "Official mistralai/Voxtral-Small-24B-2507 weights, tokenizer, and bundled model code per the pinned model card" } ], "access_boundary": "This record represents the mistralai/Voxtral-Small-24B-2507 release (July 2025). The Voxtral Mini sibling, the voxtral-small-latest alias, community quantizations and fine-tunes, and downstream products remain separate boundaries.", "strengths": [ "Ungated audio-input weights with consolidated and sharded Safetensors and tokenizer in the official repository", "The model card documents the audio length limits and Transformers and vLLM deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the Apache-2.0 grant rests on the card declaration", "The 32K token context bounds the audio input to the documented duration limits" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.74 }, "evidence": [ { "kind": "web", "label": "Official Voxtral Small 2507 weights repository and model card", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Voxtral Small model documentation", "url": "https://docs.mistral.ai/models/voxtral-small-25-07", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Voxtral launch announcement", "url": "https://mistral.ai/news/voxtral", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/mistral/voxtral-small-24b-2507.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-mixedbread-toast-1", "source_id": "mixedbread/toast-1", "name": "Toast 1", "developer": "Mixedbread AI", "url": "https://www.mixedbread.com/blog/toast-1", "description": "Mixedbread AI's Toast 1, a specialized search and knowledge-task model with tool-calling and 131K context, released August 13, 2026, served through the Mixedbread API and listed on Vercel AI Gateway.", "model_type": "language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Toast 1", "description": "Specialized search model for knowledge-intensive questions, multi-step retrieval, and evidence synthesis", "family": null, "release_date": "2026-08-13", "last_updated": "2026-08-13", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 131000, "input": null, "output": 4000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Mixedbread-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Toast 1 is served, not distributed; the governing terms are the Mixedbread service terms covering the search and API surface.", "license_evidence": [ { "license_id": "LicenseRef-Mixedbread-Service-Terms", "scope": "Toast 1 served through the Mixedbread API under Mixedbread service terms", "kind": "web_terms", "url": "https://www.mixedbread.com/pages/terms", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Toast 1 release of August 13, 2026, a search-agent specialization distinct from the open toast-harness tool loop. The Mixedbread Search service and independent hosts are separate boundaries.", "strengths": [ "Dated first-party announcement and changelog entry with full API documentation", "Dedicated model identifier with documented limits, and a second hosting path on Vercel AI Gateway" ], "tradeoffs": [ "API-only with no weights and no standalone model page beyond the agent and search surface", "The search-agent framing needs the language-model boundary kept distinct from the Mixedbread service" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.26 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/mixedbread/toast-1.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://www.mixedbread.com/blog/toast-1", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-moonshotai-kimi-k2", "source_id": null, "name": "Kimi K2 Instruct", "developer": "Moonshot AI", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "description": "A developer-published July 2025 1-trillion-parameter mixture-of-experts instruct release with block-fp8 downloadable weights governed by Moonshot AI's Modified MIT License and its usage-threshold attribution clause.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Kimi K2 Instruct", "description": "1T-parameter mixture-of-experts instruct model with 32B active parameters and a 128K context window", "family": "kimi", "release_date": null, "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 128000, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct" } ] }, "licenses": [ "LicenseRef-Kimi-K2-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Both the repository and the model weights are released under Moonshot AI's Modified MIT License, which reproduces the MIT grant but requires commercial products above 100 million monthly active users or 20 million US dollars in monthly revenue to prominently display 'Kimi K2'. That conditional attribution obligation is not part of the OSI-approved MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Kimi-K2-Modified-MIT", "scope": "moonshotai/Kimi-K2-Instruct repository code and model weights, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the moonshotai/Kimi-K2-Instruct repository release at commit fd1984e2b7a3350dbf7305fe73a4ede25c14de50 (last modified 2026-04-23), whose checkpoint is stored in block-fp8. Kimi-K2-Base, the 0905 Instruct update, the Kimi K2 Thinking line, the kimi-k2-0711-preview and turbo API variants, the Moonshot AI API, and downstream requantizations remain separate boundaries.", "strengths": [ "Moonshot AI publishes the block-fp8 checkpoint with an in-repository deployment guide naming vLLM, SGLang, KTransformers, and TensorRT-LLM", "The repository carries its own license file, a third-party notices file, and a tool-calling guide" ], "tradeoffs": [ "The Modified MIT License adds a usage-threshold attribution obligation, so the release is source-available rather than open source", "Moonshot AI's platform lists kimi-k2-0711-preview as a deprecated API model, and the model card carries no release date and now points to the 0905 update" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.56 }, "evidence": [ { "kind": "web", "label": "Official model card, block-fp8 checkpoint, and deployment guidance", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Kimi platform model list showing kimi-k2-0711-preview among deprecated API models", "url": "https://platform.kimi.ai/docs/models", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-moonshotai-kimi-k2-0905", "source_id": null, "name": "Kimi K2 Instruct 0905", "developer": "Moonshot AI", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905", "description": "A developer-published September 2025 update of Kimi K2 with a 256K context window and block-fp8 downloadable weights governed by Moonshot AI's Modified MIT License and its usage-threshold attribution clause.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Kimi K2 Instruct 0905", "description": "Updated 1T-parameter mixture-of-experts instruct model with 32B active parameters, a 256K context window, and stronger agentic coding", "family": "kimi", "release_date": null, "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 256000, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905" } ] }, "licenses": [ "LicenseRef-Kimi-K2-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Both the repository and the model weights are released under Moonshot AI's Modified MIT License, which reproduces the MIT grant but requires commercial products above 100 million monthly active users or 20 million US dollars in monthly revenue to prominently display 'Kimi K2'. That conditional attribution obligation is not part of the OSI-approved MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Kimi-K2-Modified-MIT", "scope": "moonshotai/Kimi-K2-Instruct-0905 repository code and model weights, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905/blob/ac6c49f04883bd0a0598b790693a72061c676629/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the moonshotai/Kimi-K2-Instruct-0905 repository release at commit ac6c49f04883bd0a0598b790693a72061c676629 (last modified 2026-01-30), whose checkpoint is stored in block-fp8. The original Kimi-K2-Instruct repository, the Kimi K2 Thinking line, the kimi-k2-0905-preview and turbo API variants, the Moonshot AI API, and downstream requantizations remain separate boundaries.", "strengths": [ "Moonshot AI publishes the block-fp8 checkpoint with an in-repository deployment guide naming vLLM, SGLang, KTransformers, and TensorRT-LLM", "The repository carries its own license file, a third-party notices file, and a tool-calling guide" ], "tradeoffs": [ "The Modified MIT License adds a usage-threshold attribution obligation, so the release is source-available rather than open source", "Moonshot AI's platform lists kimi-k2-0905-preview as a deprecated API model, and the model card carries no release date or versioning statement" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.56 }, "evidence": [ { "kind": "web", "label": "Official model card, block-fp8 checkpoint, and deployment guidance", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Kimi platform model list showing kimi-k2-0905-preview among deprecated API models", "url": "https://platform.kimi.ai/docs/models", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-moonshotai-kimi-k2-thinking", "source_id": "moonshotai/kimi-k2-thinking", "name": "Kimi K2 Thinking", "developer": "Moonshot AI", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "description": "A developer-published reasoning language-model release with a native INT4 downloadable checkpoint governed by Moonshot AI's Modified MIT License and its usage-threshold attribution clause.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Kimi K2 Thinking", "description": "Thinking Kimi model for slower research passes, planning, and hard technical questions", "family": "kimi-thinking", "release_date": "2025-11-06", "last_updated": "2025-11-06", "knowledge_cutoff": "2024-08", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking" } ] }, "licenses": [ "LicenseRef-Kimi-K2-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Both the repository and the model weights are released under Moonshot AI's Modified MIT License, which reproduces the MIT grant but requires commercial products above 100 million monthly active users or 20 million US dollars in monthly revenue to prominently display 'Kimi K2'. That conditional attribution obligation is not part of the OSI-approved MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Kimi-K2-Modified-MIT", "scope": "moonshotai/Kimi-K2-Thinking repository code and model weights, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking/blob/a51ccc050d73dab088bf7b0e2dd9b30ae85a4e55/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the moonshotai/Kimi-K2-Thinking repository release at commit a51ccc050d73dab088bf7b0e2dd9b30ae85a4e55 (last modified 2026-01-30), whose checkpoint is natively INT4. Kimi-K2-Instruct, Kimi-K2-Base, the 0905 Instruct update, the turbo API variant, the Moonshot AI API, and downstream requantizations remain separate boundaries.", "strengths": [ "Moonshot AI publishes the INT4 compressed-tensors checkpoint with an in-repository deployment guide covering vLLM, SGLang, and KTransformers configurations", "The repository carries its own license file, a third-party notices file, and a tool-calling guide" ], "tradeoffs": [ "The Modified MIT License adds a usage-threshold attribution obligation, so the release is source-available rather than open source", "Moonshot AI's platform lists kimi-k2-thinking as a deprecated API model, and the model card carries no release date or versioning statement" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.56 }, "evidence": [ { "kind": "web", "label": "Official model card, INT4 checkpoint, and deployment guidance", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Kimi platform model list showing kimi-k2-thinking among deprecated API models", "url": "https://platform.kimi.ai/docs/models", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/moonshotai/kimi-k2-thinking.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-moonshotai-kimi-k2-5", "source_id": "moonshotai/kimi-k2.5", "name": "Kimi K2.5", "developer": "Moonshot AI", "url": "https://huggingface.co/moonshotai/Kimi-K2.5", "description": "Moonshot AI's native multimodal reasoning release with downloadable compressed-tensors weights under the Modified MIT License carrying the Kimi K2.5 attribution clause.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Kimi K2.5", "description": "Earlier Kimi frontier model for long-context agents, coding, and multimodal work", "family": "kimi-k2", "release_date": "2026-01", "last_updated": "2026-01", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2.5" } ] }, "licenses": [ "LicenseRef-Kimi-K2-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Both the code repository and the model weights are released under the Modified MIT License, which reproduces the MIT grant but requires commercial products above 100 million monthly active users or 20 million US dollars in monthly revenue to prominently display 'Kimi K2.5'. That conditional attribution obligation is not part of the OSI-approved MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Kimi-K2-Modified-MIT", "scope": "moonshotai/Kimi-K2.5 repository code and model weights, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/moonshotai/Kimi-K2.5/blob/4d01dfe0332d63057c186e0b262165819efb6611/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the moonshotai/Kimi-K2.5 weight release at commit 4d01dfe0332d63057c186e0b262165819efb6611 (last modified 2026-04-30). The Kimi-K2-Thinking predecessor, the Kimi-K2.6 and Kimi-K2.7-Code successors, the Moonshot AI API — which discontinued kimi-k2.5 on August 31, 2026 — and downstream requantizations remain separate boundaries.", "strengths": [ "Moonshot AI publishes 64 compressed-tensors shards with the license file, third-party notices, and an in-repository deployment guide", "The repository carries modeling code, processor code, and chat-template tool-calling support beside the weights" ], "tradeoffs": [ "The Modified MIT License adds a usage-threshold attribution obligation, so the release is source-available rather than open source", "Moonshot AI discontinued the kimi-k2.5 API endpoint on August 31, 2026, leaving self-operation as the supported path" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.56 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and in-repository deployment guidance", "url": "https://huggingface.co/moonshotai/Kimi-K2.5", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Moonshot AI model list: kimi-k2.5 discontinued August 31, 2026", "url": "https://platform.moonshot.ai/docs/models", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/moonshotai/kimi-k2.5.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-moonshotai-kimi-k2-6", "source_id": "moonshotai/kimi-k2.6", "name": "Kimi K2.6", "developer": "Moonshot AI", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "description": "Moonshot AI's native multimodal agentic release with downloadable weights under the Modified MIT License carrying the Kimi K2.6 attribution clause.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Kimi K2.6", "description": "Multimodal Kimi workhorse for agent loops, coding tasks, and visual context", "family": "kimi-k2", "release_date": "2026-04-21", "last_updated": "2026-04-21", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2.6" } ] }, "licenses": [ "LicenseRef-Kimi-K2-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Both the code repository and the model weights are released under the Modified MIT License, which reproduces the MIT grant but requires commercial products above 100 million monthly active users or 20 million US dollars in monthly revenue to prominently display 'Kimi K2.6'. That conditional attribution obligation is not part of the OSI-approved MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Kimi-K2-Modified-MIT", "scope": "moonshotai/Kimi-K2.6 repository code and model weights, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the moonshotai/Kimi-K2.6 weight release at commit 7eb5002f6aadc958aed6a9177b7ed26bb94011bb (last modified 2026-05-19). The Kimi-K2.5 predecessor, the Kimi-K2.7-Code and Kimi-K3 successors, the platform API, and downstream requantizations remain separate boundaries.", "strengths": [ "Moonshot AI publishes 64 compressed-tensors shards with the license file, evaluation results, and an in-repository deployment guide", "The first-party platform lists kimi-k2.6 as a served model with text, image, and video input at 256K context" ], "tradeoffs": [ "The Modified MIT License adds a usage-threshold attribution obligation, so the release is source-available rather than open source", "Dated April–May 2026 commit trail with no retirement posture, and no per-snapshot versioning statement" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, evaluation results, and deployment guidance", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Moonshot AI model list naming kimi-k2.6", "url": "https://platform.moonshot.ai/docs/models", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/moonshotai/kimi-k2.6.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-moonshotai-kimi-k2-7-code", "source_id": "moonshotai/kimi-k2.7-code", "name": "Kimi K2.7 Code", "developer": "Moonshot AI", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "description": "Moonshot AI's dedicated coding release with downloadable weights under the Modified MIT License carrying the Kimi K2.7 Code attribution clause.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Kimi K2.7 Code", "description": "Coding-focused Kimi model, stronger on long-horizon repo work with less overthinking", "family": "kimi-k2", "release_date": "2026-06-12", "last_updated": "2026-06-12", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code" } ] }, "licenses": [ "LicenseRef-Kimi-K2-Modified-MIT" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Both the code repository and the model weights are released under the Modified MIT License, which reproduces the MIT grant but requires commercial products above 100 million monthly active users or 20 million US dollars in monthly revenue to prominently display 'Kimi K2.7 Code'. That conditional attribution obligation is not part of the OSI-approved MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Kimi-K2-Modified-MIT", "scope": "moonshotai/Kimi-K2.7-Code repository code and model weights, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/blob/74797c9c62378b951a1f6fcf5c4631024e9b8bef/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the moonshotai/Kimi-K2.7-Code weight release at commit 74797c9c62378b951a1f6fcf5c4631024e9b8bef (last modified 2026-06-15). The Kimi-K2.6 sibling line, Kimi-K3, the kimi-k2.7-code-highspeed endpoint tier, the platform API, and downstream requantizations remain separate boundaries.", "strengths": [ "Moonshot AI publishes 64 compressed-tensors shards with the license file, third-party notices, and an in-repository deployment guide", "The first-party platform lists kimi-k2.7-code as its dedicated coding model at 256K context" ], "tradeoffs": [ "The Modified MIT License adds a usage-threshold attribution obligation, so the release is source-available rather than open source", "No changelog or retirement posture accompanies the dated release" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 6.7 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and in-repository deployment guidance", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official Moonshot AI model list naming kimi-k2.7-code", "url": "https://platform.moonshot.ai/docs/models", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/moonshotai/kimi-k2.7-code.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16" }, { "id": "model-moonshotai-kimi-k2-8-preview", "source_id": "moonshotai/kimi-k2.8-preview", "name": "Kimi K2.8 Preview", "developer": "Moonshot AI", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "description": "Moonshot AI's Kimi K2.8 Preview, a text, image, and video reasoning model with 1M-token context released September 11, 2026, served behind the kimi-for-coding API identifier across Kimi Code surfaces.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Kimi K2.8 Preview", "description": "Kimi coding model with more efficient thinking and up to 1M context, available through Kimi Code", "family": "kimi-k2", "release_date": "2026-09-11", "last_updated": "2026-09-11", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 1048576, "input": null, "output": null }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Kimi K2.8 Preview is served, not distributed, under the Moonshot platform and Kimi Code service terms; no weights are released and no existing Kimi weight-license identifier applies to this API-only preview.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "Kimi K2.8 Preview served through Kimi Code under Moonshot platform service terms", "kind": "web_terms", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Kimi K2.8 Preview release of September 11, 2026, served behind the unchanged kimi-for-coding API identifier across Kimi Code, Kimi Code CLI, and Kimi Work. The reviewed Kimi K3, K2.7 Code, and earlier K2 records are separate boundaries.", "strengths": [ "Dated first-party changelog identity and a full rollout across all Kimi membership tiers", "Distinct capabilities including 1M-token context and K3-style thinking controls versus K2.7 Code" ], "tradeoffs": [ "Identity is a changelog entry only, with no dedicated model page, and the release serves behind a rotating alias", "Preview label and no weights, so lifecycle and artifact certainty are limited" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.88 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/moonshotai/kimi-k2.8-preview.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-moonshotai-kimi-k3", "source_id": "moonshotai/kimi-k3", "name": "Kimi K3", "developer": "Moonshot AI", "url": "https://huggingface.co/moonshotai/Kimi-K3", "description": "A developer-published native-multimodal agentic model release with downloadable MXFP4 weights under the bespoke Kimi K3 License, which gates model-as-a-service commercial use on revenue and adds an attribution clause.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Kimi K3", "description": "Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work", "family": "kimi-k3", "release_date": "2026-07-16", "last_updated": "2026-07-16", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Kimi-K3-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "Both the code repository and the model weights are released under the Kimi K3 License, an in-repository license file reproducing an MIT-style grant expanded to weights, parameters, configs, and inference and training code. It requires operators of a model-as-a-service business above 20 million US dollars in aggregate affiliate revenue over any consecutive twelve months to sign a separate agreement before any commercial use, and requires prominent Kimi K3 attribution above 100 million monthly active users or 20 million US dollars in monthly revenue. Those commercial gates go beyond the Kimi K2 Modified MIT text, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Kimi-K3-License", "scope": "moonshotai/Kimi-K3 repository code, model weights, parameters, configuration files, inference and training code, and documentation, including derivative works", "kind": "web_terms", "url": "https://huggingface.co/moonshotai/Kimi-K3/blob/f831ab66814297da540d832a5235f8e904f29d06/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the moonshotai/Kimi-K3 repository release with native MXFP4 weights. The Kimi K2 family, Kimi K2.5, K2.6, and K2.7-Code, the Kimi API and platform products, and downstream requantizations remain separate boundaries.", "strengths": [ "Developer-published full weights with modeling and processor code and an in-repository license file, plus vLLM, SGLang, and TokenSpeed deployment recipes", "The first-party platform quickstart documents kimi-k3 API access with vision, reasoning-effort, and tool-calling guides, and Together AI documents independent serverless hosting" ], "tradeoffs": [ "The Kimi K3 License adds a model-as-a-service revenue gate requiring a separate agreement before any commercial use, so the weights are source-available rather than open source", "Practical self-operation targets supernode-class hardware and requires trusted remote code" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.88 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and deployment guidance", "url": "https://huggingface.co/moonshotai/Kimi-K3", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Kimi K3 release blog", "url": "https://www.kimi.com/blog/kimi-k3", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Kimi API quickstart for the kimi-k3 model", "url": "https://platform.kimi.ai/docs/guide/kimi-k3-quickstart", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Together AI quickstart documenting third-party hosting", "url": "https://docs.together.ai/docs/kimi-k3-quickstart", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/moonshotai/kimi-k3.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-motif-technologies-motif-3", "source_id": "motif-technologies/motif-3", "name": "Motif 3", "developer": "Motif Technologies", "url": "https://huggingface.co/Motif-Technologies/Motif-3", "description": "Motif Technologies' Motif 3, a 315-billion-parameter MoE language model with 13.2B active parameters, published as ungated open weights under a confirmed MIT license with a bundled technical report.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Motif 3", "description": "Motif 3 is a large-scale, decoder-only Mixture-of-Experts (MoE) language model with 314 billion total parameters and 13.2 billion parameters activated per token.", "family": "motif", "release_date": "2026-08-07", "last_updated": "2026-08-07", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": "MIT", "links": [], "weights": [ { "label": "Model weights", "url": "https://huggingface.co/Motif-Technologies/Motif-3", "format": "safetensors" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository carries a LICENSE.md file with the standard MIT text under Copyright 2026 Motif Technologies Corp., confirming the reported MIT grant.", "license_evidence": [ { "license_id": "MIT", "scope": "Motif-3 repository, Safetensors weights, tokenizer, and chat template under the MIT license file", "kind": "web_terms", "url": "https://huggingface.co/Motif-Technologies/Motif-3/blob/main/LICENSE.md", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Motif-3 weight release. The Beta and NVFP4 variants, base checkpoints, and independent hosts are separate boundaries.", "strengths": [ "Truly permissive MIT license confirmed by an in-repository license file", "First Motif line for the Atlas with a flagship-scale MoE and a first-party technical report" ], "tradeoffs": [ "Roughly 630 GB of BF16 weights make deployment heavy", "A young vendor with thin provenance and a text-only English and Korean surface" ], "score_profile": "model_access", "score": { "license_clarity": 10, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 7.26 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/motif-technologies/motif-3.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/Motif-Technologies/Motif-3", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-nousresearch-hermes-4-405b", "source_id": "nousresearch/hermes-4-405b", "name": "Hermes 4 405B", "developer": "Nous Research", "url": "https://huggingface.co/NousResearch/Hermes-4-405B", "description": "Nous Research's August 2025 Hermes 4 hybrid-reasoning release at 405B parameters, fine-tuned from Llama-3.1-405B with toggleable think chains, tool use, and 131K context, published as open weights under the Llama 3.1 Community License with its acceptable-use policy. No developer API serves this exact model. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Hermes 4 405B", "description": "Reasoning model for deliberate analysis, multi-step problem solving, and tool use", "family": "hermes", "release_date": "2025-08-26", "last_updated": "2025-08-26", "knowledge_cutoff": "2024-08-31", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": false, "structured_output": false, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 117964 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Llama-3.1-Community", "LicenseRef-Llama-3.1-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The NousResearch/Hermes-4-405B repository declares license llama3 in its model metadata and ships no LICENSE file of its own, so the grant is Meta's Llama 3.1 Community License with its incorporated acceptable-use policy: attribution and redistribution duties, a Built with Llama display requirement, a 700 million monthly-active-user commercial threshold, and prohibited-use categories, making the release source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Llama-3.1-Community", "scope": "Hermes-4-405B weights derived from Llama-3.1-405B, licensed under Meta's Llama 3.1 Community License as declared by the repository metadata", "kind": "web_terms", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE", "verified_at": "2026-10-13" }, { "license_id": "LicenseRef-Llama-3.1-AUP", "scope": "Acceptable-use policy incorporated by reference into the Llama 3.1 Community License", "kind": "web_terms", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/USE_POLICY.md", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the Hermes-4-405B flagship of the Hermes 4 family. The 14B and 70B siblings, the Hermes 4.3 line, the Hermes 3 line, the base Llama-3.1-405B weights, the Hermes Agent system, and Portal-hosted agents are separate boundaries.", "strengths": [ "Hybrid reasoning with toggleable think chains, tool use, and structured outputs documented in the model card and a dedicated technical report", "Versioned 405B open weights with safetensors manifests on the developer's own Hugging Face organization" ], "tradeoffs": [ "The Llama 3.1 grant bars use above the monthly-active-user threshold without Meta's permission and names prohibited uses, so commercial reuse needs legal review", "The 405B dense checkpoint needs multi-accelerator serving hardware, and no developer API serves it, so running it is entirely self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 4, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 5.28 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/59aa52dbc09727fa6ead433b0285970ebc05ba82/models/nousresearch/hermes-4-405b.toml", "verified_at": "2026-10-13" }, { "kind": "web", "label": "NousResearch Hermes-4-405B weight repository: llama3 licence, text generation, hybrid reasoning", "url": "https://huggingface.co/NousResearch/Hermes-4-405B", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Hermes 4 Technical Report on arXiv", "url": "https://arxiv.org/abs/2508.18255", "verified_at": "2026-10-13" }, { "kind": "web", "label": "Meta Llama downloads page: canonical Llama 3.1 licence and use-policy source", "url": "https://www.llama.com/llama-downloads/", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/NousResearch/Hermes-4-405B", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-nvidia-llama-3-1-nemotron-70b-instruct", "source_id": "nvidia/llama-3.1-nemotron-70b-instruct", "name": "Llama 3.1 Nemotron 70B Instruct", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct", "description": "An ungated 70B text instruction model aligned on reward benchmarks and tool use, governed by the NVIDIA Open Model License over the Llama 3.1 base.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Llama 3.1 Nemotron 70B Instruct", "description": "Nemotron model for efficient reasoning, coding, and specialized AI agents", "family": "nemotron", "release_date": "2025-04-15", "last_updated": "2025-04-15", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model", "LicenseRef-Llama-3.1-Community", "LicenseRef-Llama-3.1-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, worldwide, non-exclusive, royalty-free but revocable right to reproduce, modify, and distribute the work with notice obligations and guardrail and indemnity clauses, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.download.nvidia.com/licenses/nvidia-open-model-license-agreement-june-2024.pdf", "scope": "Llama-3.1-Nemotron-70B-Instruct weights and bundled model code, June 2024 edition of the NVIDIA Open Model License" }, { "license_id": "LicenseRef-Llama-3.1-Community", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_1/license/", "scope": "Llama 3.1 Community License named on the NVIDIA model card as the base-model agreement" }, { "license_id": "LicenseRef-Llama-3.1-AUP", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "scope": "Acceptable-use policy of the base Llama 3.1 model named on the NVIDIA model card" } ], "access_boundary": "This record represents the nvidia/Llama-3.1-Nemotron-70B-Instruct v1.0 release of October 2024 at Hugging Face commit a83af1f4968437064635f6726fb745e5b615e863. The Transformers-format sibling repository, the Ultra and Super variants, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated weights with the governing NVIDIA Open Model License text pinned", "The model card names the base Llama 3.1 agreement and its acceptable-use policy" ], "tradeoffs": [ "The weights ship in NeMo format, so Transformers users must convert from the sibling repository", "The trial endpoint cites the Llama terms rather than the NVIDIA license, a documented surface conflict" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 5.96 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/llama-3_1-nemotron-70b-instruct", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/llama-3.1-nemotron-70b-instruct.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-llama-3-1-nemotron-safety-guard-8b-v3", "source_id": "nvidia/llama-3.1-nemotron-safety-guard-8b-v3", "name": "Llama 3.1 Nemotron Safety Guard 8B v3", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-Safety-Guard-8B-v3", "description": "An ungated 8B content-safety guard model over the Llama 3.1 8B base that emits safe/unsafe verdicts as generated text, governed by the NVIDIA Open Model License.", "model_type": "judgment_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Llama 3.1 Nemotron Safety Guard 8B v3", "description": "Safety model for policy screening, moderation, and risk-aware routing workflows", "family": "nemotron", "release_date": "2025-10-28", "last_updated": "2025-10-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model", "LicenseRef-Llama-3.1-Community", "LicenseRef-Llama-3.1-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, the October 2025 edition whose grant is revocable with guardrail and indemnity clauses, over the Llama 3.1 Community License base named as \"Built with Llama\".", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "scope": "Llama-3.1-Nemotron-Safety-Guard-8B-v3 weights, LoRA adapter, and bundled model code, October 2025 edition of the NVIDIA Open Model License" }, { "license_id": "LicenseRef-Llama-3.1-Community", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_1/license/", "scope": "Llama 3.1 Community License named on the NVIDIA model card as the base-model agreement" }, { "license_id": "LicenseRef-Llama-3.1-AUP", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "scope": "Acceptable-use policy of the base Llama 3.1 model named on the NVIDIA model card" } ], "access_boundary": "This record represents the nvidia/Llama-3.1-Nemotron-Safety-Guard-8B-v3 v1.0 release dated October 28, 2025 at Hugging Face commit 8fdc246ba3d56db9c469d534233b9f582d3afafa. The Llama 3.1 8B base model, the v2 and earlier guard generations, downstream fine-tunes, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated weights with the LoRA adapter shipping beside the full checkpoint", "The model card documents the verdict JSON schema and Model Card++ subcards" ], "tradeoffs": [ "The revocable NVIDIA Open Model License carries guardrail and indemnity clauses", "The guard context is limited to 8K tokens by the base-model configuration" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.82 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-Safety-Guard-8B-v3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/llama-3_1-nemotron-safety-guard-8b-v3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/llama-3.1-nemotron-safety-guard-8b-v3.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-llama-3-1-nemotron-ultra-253b", "source_id": "nvidia/llama-3.1-nemotron-ultra-253b", "name": "Llama 3.1 Nemotron Ultra 253B", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1", "description": "An ungated 253B text reasoning model with a DecILM sparse architecture, governed by the NVIDIA Open Model License over the Llama 3.1 base.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Llama 3.1 Nemotron Ultra 253B", "description": "Flagship Nemotron model for high-throughput reasoning and complex agents", "family": "nemotron", "release_date": "2025-04-07", "last_updated": "2025-04-07", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model", "LicenseRef-Llama-3.1-Community", "LicenseRef-Llama-3.1-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, worldwide, non-exclusive, royalty-free but revocable right to reproduce, modify, and distribute the work with notice obligations and guardrail and indemnity clauses, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "scope": "Llama-3_1-Nemotron-Ultra-253B-v1 weights, tokenizer, and bundled model code, October 2025 edition of the NVIDIA Open Model License" }, { "license_id": "LicenseRef-Llama-3.1-Community", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_1/license/", "scope": "Llama 3.1 Community License named on the NVIDIA model card as the base-model agreement" }, { "license_id": "LicenseRef-Llama-3.1-AUP", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_1/use-policy/", "scope": "Acceptable-use policy of the base Llama 3.1 model named on the NVIDIA model card" } ], "access_boundary": "This record represents the nvidia/Llama-3_1-Nemotron-Ultra-253B-v1 v1.0 release dated April 7, 2025 at Hugging Face commit 5b47def5b8957142d82faeb44bfad9349d02840d. The 70B Instruct and Super siblings, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors shards with a consistent NVIDIA Open Model License citation on card and endpoint", "Dated release with a model version, trained-dates range, and build.nvidia.com documentation" ], "tradeoffs": [ "The DecILM architecture requires trust-remote-code loading and NVIDIA-hardware-oriented runtime documentation", "The NVIDIA-specific license is revocable with guardrail and indemnity clauses" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.62 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/llama-3_1-nemotron-ultra-253b-v1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/llama-3.1-nemotron-ultra-253b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-llama-3-3-nemotron-super-49b-v1", "source_id": "nvidia/llama-3.3-nemotron-super-49b-v1", "name": "Llama 3.3 Nemotron Super 49B v1", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1", "description": "An ungated 49B text reasoning model with a DecILM sparse architecture, governed by the NVIDIA Open Model License over the Llama 3.3 base.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Llama 3.3 Nemotron Super 49B v1", "description": "Nemotron model for efficient reasoning, coding, and specialized AI agents", "family": "nemotron", "release_date": "2025-04-07", "last_updated": "2025-04-07", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model", "LicenseRef-Llama-3.3-Community", "LicenseRef-Llama-3.3-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, worldwide, non-exclusive, royalty-free but revocable right to reproduce, modify, and distribute the work with notice obligations and guardrail and indemnity clauses, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "scope": "Llama-3_3-Nemotron-Super-49B-v1 weights, tokenizer, and bundled model code, October 2025 edition of the NVIDIA Open Model License" }, { "license_id": "LicenseRef-Llama-3.3-Community", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_3/license/", "scope": "Llama 3.3 Community License named on the NVIDIA model card as the base-model agreement" }, { "license_id": "LicenseRef-Llama-3.3-AUP", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_3/use-policy/", "scope": "Acceptable-use policy of the base Llama 3.3 model named on the NVIDIA model card" } ], "access_boundary": "This record represents the nvidia/Llama-3_3-Nemotron-Super-49B-v1 v1.0 release dated March 18, 2025 at Hugging Face commit 387156d8d6868c19f3472fa607aa9bfc4f662333. The v1.5 upgrade, the 70B Instruct and Ultra siblings, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors shards with a consistent NVIDIA Open Model License citation on card and endpoint", "Dated release with a model version and Transformers and vLLM runtime documentation" ], "tradeoffs": [ "The DecILM architecture requires trust-remote-code loading and NVIDIA-hardware-oriented runtime documentation", "The NVIDIA-specific license is revocable with guardrail and indemnity clauses" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.62 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/llama-3_3-nemotron-super-49b-v1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/llama-3.3-nemotron-super-49b-v1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-llama-3-3-nemotron-super-49b-v1-5", "source_id": "nvidia/llama-3.3-nemotron-super-49b-v1.5", "name": "Llama 3.3 Nemotron Super 49B v1.5", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5", "description": "An ungated 49B text reasoning model upgrading the v1 release with improved instruction following and tool calling, governed by the NVIDIA Open Model License over the Llama 3.3 base.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Llama 3.3 Nemotron Super 49B v1.5", "description": "Nemotron model for efficient reasoning, coding, and specialized AI agents", "family": "nemotron", "release_date": "2025-07-25", "last_updated": "2025-07-25", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model", "LicenseRef-Llama-3.3-Community", "LicenseRef-Llama-3.3-AUP" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, worldwide, non-exclusive, royalty-free but revocable right to reproduce, modify, and distribute the work with notice obligations and guardrail and indemnity clauses, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "scope": "Llama-3_3-Nemotron-Super-49B-v1_5 weights, tokenizer, and bundled model code, October 2025 edition of the NVIDIA Open Model License" }, { "license_id": "LicenseRef-Llama-3.3-Community", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_3/license/", "scope": "Llama 3.3 Community License named on the NVIDIA model card as the base-model agreement" }, { "license_id": "LicenseRef-Llama-3.3-AUP", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://developer.meta.com/ai/llama3_3/use-policy/", "scope": "Acceptable-use policy of the base Llama 3.3 model named on the NVIDIA model card" } ], "access_boundary": "This record represents the nvidia/Llama-3_3-Nemotron-Super-49B-v1_5 v1.5 release dated July 25, 2025 at Hugging Face commit 420ba7d28211abf116b8b103ab700d92619daf98. The v1 release, the 70B Instruct and Ultra siblings, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors shards with dual dated release records on Hugging Face and build.nvidia.com", "The model card documents the explicit upgrade relationship to the v1 release and a tool-call parser" ], "tradeoffs": [ "The DecILM architecture requires trust-remote-code loading and NVIDIA-hardware-oriented runtime documentation", "The NVIDIA-specific license is revocable with guardrail and indemnity clauses" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 6.74 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/llama-3_3-nemotron-super-49b-v1_5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/llama-3.3-nemotron-super-49b-v1.5.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-llama-nemotron-embed-1b-v2", "source_id": "nvidia/llama-nemotron-embed-1b-v2", "name": "Llama Nemotron Embed 1B v2", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/llama-nemotron-embed-1b-v2", "description": "NVIDIA's October 2025 second-generation Llama Nemotron embedding release at 1B parameters, trained on a commercial-use dataset blend for passage retrieval, published as open weights with binaries under the NVIDIA Open Model License, code under Apache 2.0, and Llama 3.2 community terms for the base. No developer API serves this checkpoint. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "embedding_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Llama Nemotron Embed 1B v2", "description": "Multilingual text embedding model for retrieval and question answering pipelines", "family": "nemotron", "release_date": "2025-10-16", "last_updated": "2025-10-16", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text", "embedding" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 8192, "input": null, "output": 2048 }, "reported_open_weights": true, "reported_license": "nvidia-open-model-license", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/nvidia/llama-nemotron-embed-1b-v2" } ] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model", "LicenseRef-Llama-3.2-Community", "Apache-2.0" ], "source_model": "mixed_source", "license_review_status": "verified", "license_note": "The repository LICENSE file puts all binary model files under the NVIDIA Open Model License Agreement and all source code files under Apache 2.0, and the model card additionally names the Llama 3.2 Community Model License for the base. All three material licences are recorded; the mix of open and restricted grants makes the release mixed-source rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "scope": "llama-nemotron-embed-1b-v2 binary model files published in NVIDIA's Hugging Face repository", "kind": "web_terms", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "verified_at": "2026-10-13" }, { "license_id": "LicenseRef-Llama-3.2-Community", "scope": "Llama 3.2 base terms named by the model card as additional governing terms", "kind": "web_terms", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE", "verified_at": "2026-10-13" }, { "license_id": "Apache-2.0", "scope": "Repository source code files under Apache 2.0 per the repository LICENSE file", "kind": "web_terms", "url": "https://opensource.org/licenses/Apache-2.0", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the llama-nemotron-embed-1b-v2 checkpoint. The v1 generation, larger Nemotron embedding sizes, the base Llama 3.2 weights, and NVIDIA NIM or third-party hosted endpoints are separate boundaries.", "strengths": [ "Commercial-use training blend built to avoid non-commercial dataset restrictions, documented in the model card", "1B retriever that runs leanly with the licence split stated file-by-file in the repository LICENSE" ], "tradeoffs": [ "Three overlapping licences govern weights, base, and code, so commercial reuse needs legal review of each", "No developer API or managed host serves this checkpoint, so running it is entirely self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.46 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/59aa52dbc09727fa6ead433b0285970ebc05ba82/models/nvidia/llama-nemotron-embed-1b-v2.toml", "verified_at": "2026-10-13" }, { "kind": "web", "label": "NVIDIA llama-nemotron-embed-1b-v2 weight repository: licence split and commercial dataset blend", "url": "https://huggingface.co/nvidia/llama-nemotron-embed-1b-v2", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/nvidia/llama-nemotron-embed-1b-v2", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-nvidia-nemotron-3-content-safety", "source_id": "nvidia/nemotron-3-content-safety", "name": "Nemotron 3 Content Safety", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Nemotron-3-Content-Safety", "description": "An ungated V1.1 content-safety model merged into the Gemma-3-4B-it base that emits safe/unsafe verdicts as generated text, governed by the NVIDIA Nemotron Open Model License.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron 3 Content Safety", "description": "Safety model for policy screening, moderation, and risk-aware routing workflows", "family": "nemotron", "release_date": "2026-04-16", "last_updated": "2026-04-16", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Nemotron-Open-Model", "LicenseRef-Gemma-Terms-of-Use", "LicenseRef-Gemma-Prohibited-Use-Policy" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Nemotron Open Model License, whose grant is perpetual and irrevocable. The merged Gemma-3-4B-it base adds the Gemma Terms of Use and the Gemma Prohibited Use Policy as a secondary layer, named on the model card.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Nemotron-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "scope": "Nemotron-3-Content-Safety weights and bundled model code, December 2025 edition of the NVIDIA Nemotron Open Model License" }, { "license_id": "LicenseRef-Gemma-Terms-of-Use", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://ai.google.dev/gemma/terms", "scope": "Gemma Terms of Use named on the NVIDIA model card for the merged Gemma-3-4B-it base" }, { "license_id": "LicenseRef-Gemma-Prohibited-Use-Policy", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "scope": "Gemma Prohibited Use Policy named on the NVIDIA model card for the merged Gemma-3-4B-it base" } ], "access_boundary": "This record represents the nvidia/Nemotron-3-Content-Safety V1.1 release of March 2026 at Hugging Face commit b47f6f6c18dc7d11a786abbbca2510cba58ede20. The Gemma-3-4B-it base model, the 3.5 content-safety line, downstream fine-tunes, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated merged weights with Model Card++ subcards and a released training dataset", "The model card documents the verdict schema and Transformers and vLLM runtimes" ], "tradeoffs": [ "The web-hosted NVIDIA license can change independently of the pinned repository", "The dual Gemma base layer adds restricted-use terms to the governing instrument" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.04 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Nemotron-3-Content-Safety", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-3-content-safety", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-3-content-safety.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-3-nano-30b-a3b", "source_id": "nvidia/nemotron-3-nano-30b-a3b", "name": "Nemotron 3 Nano 30B A3B", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "description": "A developer-published hybrid mixture-of-experts text model whose ungated BF16 weights are governed by the NVIDIA Nemotron Open Model License rather than an OSI-approved software license.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron 3 Nano 30B A3B", "description": "Small Nemotron 3 MoE for efficient coding, math, and long-context agents", "family": "nemotron", "release_date": "2025-12-15", "last_updated": "2025-12-15", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Nemotron-Open-Model" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Nemotron Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, irrevocable, commercially usable right to reproduce, modify, and distribute the work with notice obligations, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Nemotron-Open-Model", "scope": "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 weights, tokenizer, and bundled model code", "kind": "web_terms", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 v1.0 release dated December 15, 2025. The Base-BF16, FP8, and NVFP4 repositories, the 4B and Omni variants, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "NVIDIA publishes ungated Safetensors shards with a stated model version, release date, and links to the base variant", "The model card documents Transformers, vLLM, TensorRT-LLM, SGLang, and llama.cpp paths and ships the reasoning parser and model code in the repository" ], "tradeoffs": [ "The governing license lives on an NVIDIA web page that can change independently of the pinned repository", "Loading requires trusted remote code from the repository and the documented hardware list is NVIDIA-only" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.04 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and inference-engine guidance", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "verified_at": "2026-09-11" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-3-nano-30b-a3b", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/nvidia/nemotron-3-nano-30b-a3b.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-nvidia-nemotron-3-nano-omni-30b-a3b-reasoning", "source_id": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning", "name": "Nemotron 3 Nano Omni 30B A3B Reasoning", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16", "description": "An ungated 30B multimodal reasoning model handling text, image, video, and audio input, governed by the NVIDIA Open Model Agreement.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron 3 Nano Omni 30B A3B Reasoning", "description": "Open Nemotron omni model combining reasoning with text, vision, and audio", "family": "nemotron", "release_date": "2026-04-28", "last_updated": "2026-04-28", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model-Agreement" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card and the trial endpoint both state that use of the model is governed by the NVIDIA Open Model Agreement, a web-hosted NVIDIA document distinct from the Nemotron Open Model License and the Open Model License used by other Nemotron releases. Its grant is perpetual, worldwide, non-exclusive, no-charge, royalty-free, and irrevocable, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model-Agreement", "scope": "Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 weights, tokenizer, and bundled model code", "kind": "web_terms", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-agreement/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 v1.0 release dated April 28, 2026 at Hugging Face commit e5e9932441de940c9a62185c870ea5bcd4cd24e2. The text-only Nano sibling, the FP8 and NVFP4 variants, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors weights with a consistent NVIDIA Open Model Agreement citation on card and endpoint", "Dated release with video, audio, image, and text input support and a technical report" ], "tradeoffs": [ "The NVIDIA-specific license is not an OSI-approved instrument", "Reasoning is enabled by default, which raises per-request inference cost" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.04 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-3-super-120b-a12b", "source_id": "nvidia/nemotron-3-super-120b-a12b", "name": "Nemotron 3 Super 120B A12B", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "description": "An ungated 120B text reasoning model with an A12B mixture-of-experts architecture, governed by the NVIDIA Nemotron Open Model License.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron 3 Super 120B A12B", "description": "Nemotron middle tier for collaborative agents and high-volume reasoning workloads", "family": "nemotron", "release_date": "2026-03-11", "last_updated": "2026-03-11", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Nemotron-Open-Model" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Nemotron Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable right to reproduce, modify, and distribute the work with notice obligations, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Nemotron-Open-Model", "scope": "NVIDIA-Nemotron-3-Super-120B-A12B-BF16 weights, tokenizer, and bundled model code", "kind": "web_terms", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 v1.0 release dated March 2026 at Hugging Face commit 2dc98e2afe4face0e4ce40972a915c45368bd34a. The Ultra and Nano siblings, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors weights with the governing Nemotron Open Model License cited on card and endpoint", "Dated release with a technical report and reproducibility configuration" ], "tradeoffs": [ "The NVIDIA-specific license is not an OSI-approved instrument", "Frontier-scale inference needs multi-GPU serving with NVIDIA-oriented runtime documentation" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.04 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-3-super-120b-a12b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-3-ultra-550b-a55b", "source_id": "nvidia/nemotron-3-ultra-550b-a55b", "name": "Nemotron 3 Ultra 550B A55B", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "description": "An ungated 550B text reasoning model with an A55B mixture-of-experts architecture and up to 1M-token context, governed by the OpenMDW License Agreement 1.1.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron 3 Ultra 550B A55B", "description": "Largest Nemotron 3 model for maximum open-weight reasoning and agent accuracy", "family": "nemotron", "release_date": "2026-06-04", "last_updated": "2026-06-04", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 128000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card and the trial endpoint both state that use of the model is governed by the OpenMDW License Agreement 1.1, a permissive-style third-party instrument that imposes no use restrictions beyond notice and litigation termination clauses. It is not OSI-approved, so the record classifies it as open source consistent with the taxonomy classification of the instrument.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 weights, tokenizer, and bundled model code", "kind": "web_terms", "url": "https://raw.githubusercontent.com/OpenMDW/OpenMDW/refs/heads/main/1.1/LICENSE.OpenMDW-1.1", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 v1.0 release dated June 4, 2026 at Hugging Face commit 77df655d5e9f8362164ed14dd8b48f8bce657498. The Super and Nano siblings, the NVFP4 checkpoint served by the trial endpoint, downstream quantizations, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors weights with a consistent OpenMDW-1.1 citation on card, metadata, and trial endpoint", "Dated release with model dates, cutoffs, and a technical report" ], "tradeoffs": [ "Frontier-scale footprint needs eight or more B200-class or 16 H100 GPUs with multi-node orchestration", "The trial endpoint serves the NVFP4 variant rather than the BF16 release under review" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.84 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-3-ultra-550b-a55b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-3-5-content-safety", "source_id": "nvidia/nemotron-3.5-content-safety", "name": "Nemotron 3.5 Content Safety", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Nemotron-3.5-Content-Safety", "description": "An ungated V1.2 content-safety model merged into the Gemma-3-4B-it base that emits safe/unsafe verdicts as generated text, governed by the OpenMDW License Agreement 1.1.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron 3.5 Content Safety", "description": "Safety model for policy screening, moderation, and risk-aware routing workflows", "family": "nemotron", "release_date": "2026-06-04", "last_updated": "2026-06-04", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the OpenMDW License Agreement 1.1, a permissive instrument classified as open source in the taxonomy. The merged Gemma-3-4B-it base also carries the Gemma Terms of Use and the Gemma Prohibited Use Policy as a secondary layer, named on the model card.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "Nemotron-3.5-Content-Safety weights and bundled model code", "kind": "web_terms", "url": "https://raw.githubusercontent.com/OpenMDW/OpenMDW/refs/heads/main/1.1/LICENSE.OpenMDW-1.1", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the nvidia/Nemotron-3.5-Content-Safety V1.2 release of June 2026 at Hugging Face commit 35645ed3543b7e7ffaed2e788699e57a5051497c. The Gemma-3-4B-it base model, the Nemotron 3 content-safety line, downstream fine-tunes, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated merged weights with the training dataset released and an Ollama Modelfile shipped", "The model card documents the verdict schema with optional reasoning traces and custom-policy modes" ], "tradeoffs": [ "The merged Gemma base layer adds restricted-use terms beside the permissive OpenMDW instrument", "The NVIDIA blog's stale license prose names the Open Model License while the card governs" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.46 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Nemotron-3.5-Content-Safety", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-3.5-content-safety", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-3.5-content-safety.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-3-5-lightning", "source_id": "nvidia/nemotron-3.5-lightning", "name": "Nemotron 3.5 Lightning 30B A3B", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "description": "An ungated 30B text reasoning model with an A3B mixture-of-experts architecture and up to 1M-token context, governed by the OpenMDW License Agreement 1.1.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron 3.5 Lightning 30B A3B", "description": "Fast NVIDIA Nemotron MoE for reliable agentic tasks across enterprise workloads", "family": "nemotron", "release_date": "2026-08-11", "last_updated": "2026-08-11", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 262144 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships the OpenMDW License Agreement 1.1 file, and the card and trial endpoint both name it as the governing instrument. It imposes no use restrictions beyond notice and litigation termination clauses but is not OSI-approved, so the record classifies it as open source consistent with the taxonomy classification of the instrument.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 weights, tokenizer, and bundled model code", "kind": "web_terms", "url": "https://raw.githubusercontent.com/OpenMDW/OpenMDW/refs/heads/main/1.1/LICENSE.OpenMDW-1.1", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 v1.0 release dated August 11, 2026 at Hugging Face commit a9904d24bcc1d289a1950fa9d2b978c47cf903b9. The NVFP4 checkpoint served by the trial endpoint, the 3.x generation, downstream quantizations, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors weights with the OpenMDW-1.1 license file shipped in the repository", "Dated GA release with model dates, validated 256K context, and technical documentation" ], "tradeoffs": [ "The permissive OpenMDW-1.1 instrument is not OSI-approved", "The trial endpoint serves the NVFP4 variant rather than the BF16 release under review" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 7.38 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-3.5-lightning-30b-a3b", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-3.5-lightning.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-cascade-2-30b-a3b", "source_id": "nvidia/nemotron-cascade-2-30b-a3b", "name": "Nemotron Cascade 2 30B A3B", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B", "description": "An ungated 30B text reasoning model post-trained with Cascade reinforcement learning, governed by the NVIDIA Open Model License.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Nemotron Cascade 2 30B A3B", "description": "Nemotron model for efficient reasoning, coding, and specialized AI agents", "family": "nemotron", "release_date": "2026-03-24", "last_updated": "2026-04-09", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, the October 2025 edition whose grant is revocable and which adds guardrail-termination, ethics, and indemnity clauses. It is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "scope": "Nemotron-Cascade-2-30B-A3B weights, tokenizer, and bundled model code, October 2025 edition of the NVIDIA Open Model License", "kind": "web_terms", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the nvidia/Nemotron-Cascade-2-30B-A3B release dated March 2026 at Hugging Face commit 6327cdbcf907e1c7cec9cb29fb6e6cebdf8feaf7. The Cascade 1 generation, the Nemotron 3 siblings, downstream quantizations, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors weights with a technical report and solution assets", "The model card names the governing license and the training methodology" ], "tradeoffs": [ "The revocable October 2025 edition of the NVIDIA Open Model License carries guardrail and indemnity clauses", "No build.nvidia.com trial endpoint is published for this release, so reach is download-only" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.32 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-cascade-2-30b-a3b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-content-safety-reasoning-4b", "source_id": "nvidia/nemotron-content-safety-reasoning-4b", "name": "Nemotron Content Safety Reasoning 4B", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Nemotron-Content-Safety-Reasoning-4B", "description": "An ungated 4B content-safety model merged into the Gemma-3-4B-it base that emits harm verdicts with optional reasoning traces as generated text, governed by the NVIDIA Open Model License.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron Content Safety Reasoning 4B", "description": "Safety model for policy screening, moderation, and risk-aware routing workflows", "family": "nemotron", "release_date": "2026-01-22", "last_updated": "2026-01-22", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model", "LicenseRef-Gemma-Terms-of-Use", "LicenseRef-Gemma-Prohibited-Use-Policy" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, the October 2025 edition whose grant is revocable with guardrail and indemnity clauses. The merged Gemma-3-4B-it base adds the Gemma Terms of Use and the Gemma Prohibited Use Policy as a secondary layer.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "scope": "Nemotron-Content-Safety-Reasoning-4B weights and bundled model code, October 2025 edition of the NVIDIA Open Model License" }, { "license_id": "LicenseRef-Gemma-Terms-of-Use", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://ai.google.dev/gemma/terms", "scope": "Gemma Terms of Use named on the NVIDIA model card for the merged Gemma-3-4B-it base" }, { "license_id": "LicenseRef-Gemma-Prohibited-Use-Policy", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://ai.google.dev/gemma/prohibited_use_policy", "scope": "Gemma Prohibited Use Policy named on the NVIDIA model card for the merged Gemma-3-4B-it base" } ], "access_boundary": "This record represents the nvidia/Nemotron-Content-Safety-Reasoning-4B v1.0 release of November 2025 at Hugging Face commit 29da6ec41d9c710d937afcd41f65e241ab822792. The Gemma-3-4B-it base model, the Nemotron 3 content-safety lines, downstream fine-tunes, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated weights with the verdict schema, reasoning traces, and three worked examples documented", "The model card cites the EMNLP 2025 paper and releases dataset links" ], "tradeoffs": [ "Leftover DeepSpeed training-state files clutter the weight repository", "The release dates conflict between the card (November 2025) and the trial endpoint (January 2026)" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 7, "deployment_portability": 7, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.52 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Nemotron-Content-Safety-Reasoning-4B", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-content-safety-reasoning-4b", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-content-safety-reasoning-4b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-mini-4b-instruct", "source_id": "nvidia/nemotron-mini-4b-instruct", "name": "Nemotron Mini 4B Instruct", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct", "description": "An ungated 4B text instruction model for on-device and edge workloads, governed by the NVIDIA Community Model License of August 2024.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Nemotron Mini 4B Instruct", "description": "Compact Nemotron model for efficient reasoning and deployable AI agents", "family": "nemotron", "release_date": "2024-08-21", "last_updated": "2024-08-26", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Community-Model-2024" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository ships a zero-byte LICENSE placeholder beside the governing nvidia-community-model-license-aug2024.pdf, while the card front matter names the different NVIDIA Open Model License and the trial page cites the older NVIDIA AI Foundation Models Community License Agreement. The in-repo August 2024 PDF governs, with a revocable grant and a production gate on NVIDIA NIM runtimes, and the conflicting pointers open a durable license review incident.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Community-Model-2024", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct/blob/791833e92ebddb0bc2c1007f6d2b6764f886a2ae/nvidia-community-model-license-aug2024.pdf", "scope": "Nemotron-Mini-4B-Instruct weights and bundled model code, NVIDIA Community Model License dated August 23, 2024" } ], "access_boundary": "This record represents the nvidia/Nemotron-Mini-4B-Instruct release at Hugging Face commit 791833e92ebddb0bc2c1007f6d2b6764f886a2ae. The Nemotron 3 generation, downstream quantizations, the NVIDIA build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated weights with the governing NVIDIA Community Model License pinned in the repository", "The build.nvidia.com trial endpoint and NIM packaging are documented" ], "tradeoffs": [ "The license is revocable, non-sublicensable, and gates production use on an NVIDIA NIM runtime subscription", "The card front matter, the trial page, and the in-repo license file point to three different instruments" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 7, "deployment_portability": 5, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 6, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint and API terms", "url": "https://build.nvidia.com/nvidia/nemotron-mini-4b-instruct", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-mini-4b-instruct.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-nano-12b-v2-vl", "source_id": "nvidia/nemotron-nano-12b-v2-vl", "name": "Nemotron Nano 12B v2 VL", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16", "description": "An ungated 12B vision-language reasoning model handling text, image, and video input, governed by the NVIDIA Open Model License, whose trial endpoint reached end of life in August 2026.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Nemotron Nano 12B v2 VL", "description": "Nemotron multimodal model for visual reasoning and agentic AI workflows", "family": "nemotron", "release_date": "2025-10-28", "last_updated": "2025-10-28", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 128000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, worldwide, non-exclusive, royalty-free but revocable right to reproduce, modify, and distribute the work with notice obligations and guardrail and indemnity clauses, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "scope": "NVIDIA-Nemotron-Nano-12B-v2-VL-BF16 weights, tokenizer, and bundled model code, October 2025 edition of the NVIDIA Open Model License" } ], "access_boundary": "This record represents the nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16 v1.0 release dated October 28, 2025 at Hugging Face commit ca9543b126e8bf3176916d3d305ccc415f89fd4d. The FP8 and NVFP4 variants, the 9B text sibling, downstream quantizations, the retired build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors shards with a dated release, model version, and technical report", "Video and image processors ship beside the weights, and NIM container packaging persists" ], "tradeoffs": [ "The NVIDIA-specific license is revocable with guardrail and indemnity clauses", "The build.nvidia.com trial endpoint returned end-of-life in August 2026, removing the hosted trial path" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.74 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint documentation", "url": "https://build.nvidia.com/nvidia/nemotron-nano-12b-v2-vl", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-nano-12b-v2-vl.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-nvidia-nemotron-nano-9b-v2", "source_id": "nvidia/nemotron-nano-9b-v2", "name": "Nemotron Nano 9B v2", "developer": "NVIDIA", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2", "description": "An ungated 9B text reasoning model with tool-call support, governed by the NVIDIA Open Model License, whose trial endpoint reached end of life in August 2026.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Nemotron Nano 9B v2", "description": "Compact Nemotron model for efficient reasoning and deployable AI agents", "family": "nemotron", "release_date": "2025-08-18", "last_updated": "2025-08-18", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-NVIDIA-Open-Model" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card states that use of the model is governed by the NVIDIA Open Model License, a web-hosted NVIDIA document rather than a file in the repository. The license grants a perpetual, worldwide, non-exclusive, royalty-free but revocable right to reproduce, modify, and distribute the work with notice obligations and guardrail and indemnity clauses, but it is an NVIDIA-specific instrument, not an OSI-approved license.", "license_evidence": [ { "license_id": "LicenseRef-NVIDIA-Open-Model", "kind": "web_terms", "verified_at": "2026-09-18", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "scope": "NVIDIA-Nemotron-Nano-9B-v2 weights, tokenizer, and bundled model code, October 2025 edition of the NVIDIA Open Model License" } ], "access_boundary": "This record represents the nvidia/NVIDIA-Nemotron-Nano-9B-v2 v1.0 release dated August 18, 2025 at Hugging Face commit 6533e8de2c68e4536bf7c411d7a3ce5734111476. The FP8 and NVFP4 variants, the 12B VL sibling, downstream quantizations, the retired build.nvidia.com trial endpoint, and NIM packaging are separate model or service boundaries.", "strengths": [ "Ungated Safetensors shards with a dated release, model version, and technical report", "The trial endpoint's deprecation and end-of-life dates are documented, and NIM packaging persists" ], "tradeoffs": [ "The NVIDIA-specific license is revocable with guardrail and indemnity clauses", "The build.nvidia.com trial endpoint returned end-of-life in August 2026, removing the hosted trial path" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 3, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 6.86 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and runtime guidance", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "NVIDIA-hosted trial endpoint documentation", "url": "https://build.nvidia.com/nvidia/nvidia-nemotron-nano-9b-v2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/nvidia/nemotron-nano-9b-v2.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-openai-gpt-image-2-5-flare", "source_id": "openai/gpt-image-2.5-flare", "name": "GPT Image 2.5 Flare", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-image-2.5-flare", "description": "OpenAI's fastest image generation model for everyday use, accepting text and image inputs and producing image outputs, with a dated snapshot behind the undated ID, offered through OpenAI's developer API and documented Azure Foundry hosting, without downloadable weights.", "model_type": "image_generation_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT Image 2.5 Flare", "description": "Image model for prompt-driven generation, editing, and visual design workflows", "family": "gpt-image", "release_date": "2026-09-08", "last_updated": "2026-09-08", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 0, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT Image 2.5 Flare as an API service for business and developer customers under the OpenAI Services Agreement, which applies to use of OpenAI's APIs and assigns ownership of Output to the customer. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the model through OpenAI's APIs, and ownership of Output", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the GPT Image 2.5 Flare release identified by the model ID gpt-image-2.5-flare and its dated snapshot gpt-image-2.5-flare-2026-09-08. The sibling GPT Image 2.5 Sunburst tier, the earlier GPT-Image-2 release, the deprecated gpt-image-1 family, ChatGPT image features, the OpenAI API platform, Azure Foundry hosting of the same model, and fine-tuned or downstream derivatives are separate model, service, or application boundaries.", "strengths": [ "The official model page tells developers to use the undated ID or pin the dated snapshot gpt-image-2.5-flare-2026-09-08", "Microsoft's own Foundry models listing names gpt-image-2.5-flare as an image generation model, adding managed serving beyond the first-party API" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed APIs rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT Image 2.5 Flare model, modality, endpoint, and dated-snapshot documentation", "url": "https://developers.openai.com/api/docs/models/gpt-image-2.5-flare", "verified_at": "2026-10-06" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Microsoft Foundry Models sold by Azure listing naming gpt-image-2.5-flare", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official deprecations page checked: no retirement notice for this model", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official image generation guide naming Flare and Sunburst as the current OpenAI image models", "url": "https://developers.openai.com/api/docs/guides/image-generation", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/openai/gpt-image-2.5-flare.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "The results indicate that neither GPT-Image-2.5-Sunburst nor GPT-Image-2.5-Flare cross the Bio High threshold or the Cyber High threshold. We, regardless and as done in GPT-Images 2.0, take a precautionary approach and apply mitigations appropriate for a model treated as High capability in the biological-risk domain.", "url": "https://deploymentsafety.openai.com/chatgpt-images-2-5/", "content_sha256": "4707244c63d42d903e09d7a0afb42e8f9c52e6fc2da07537e49d7263df057081", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-image-2-5-sunburst", "source_id": "openai/gpt-image-2.5-sunburst", "name": "GPT Image 2.5 Sunburst", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst", "description": "OpenAI's most capable image generation and editing model, accepting text and image inputs and producing image outputs, with a dated snapshot behind the undated ID, offered through OpenAI's developer API and documented Azure Foundry hosting, without downloadable weights.", "model_type": "image_generation_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT Image 2.5 Sunburst", "description": "Image model for prompt-driven generation, editing, and visual design workflows", "family": "gpt-image", "release_date": "2026-09-08", "last_updated": "2026-09-08", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 0, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT Image 2.5 Sunburst as an API service for business and developer customers under the OpenAI Services Agreement, which applies to use of OpenAI's APIs and assigns ownership of Output to the customer. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the model through OpenAI's APIs, and ownership of Output", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the GPT Image 2.5 Sunburst release identified by the model ID gpt-image-2.5-sunburst and its dated snapshot gpt-image-2.5-sunburst-2026-09-08. The sibling GPT Image 2.5 Flare tier, the earlier GPT-Image-2 release, the deprecated gpt-image-1 family, ChatGPT image features, the OpenAI API platform, Azure Foundry hosting of the same model, and fine-tuned or downstream derivatives are separate model, service, or application boundaries.", "strengths": [ "The official model page tells developers to use the undated ID or pin the dated snapshot gpt-image-2.5-sunburst-2026-09-08", "Microsoft's own Foundry models listing names gpt-image-2.5-sunburst as an image generation model, adding managed serving beyond the first-party API" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed APIs rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT Image 2.5 Sunburst model, modality, endpoint, and dated-snapshot documentation", "url": "https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst", "verified_at": "2026-10-06" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Microsoft Foundry Models sold by Azure listing naming gpt-image-2.5-sunburst", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official deprecations page checked: no retirement notice for this model", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official image generation guide naming Flare and Sunburst as the current OpenAI image models", "url": "https://developers.openai.com/api/docs/guides/image-generation", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/openai/gpt-image-2.5-sunburst.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "The results indicate that neither GPT-Image-2.5-Sunburst nor GPT-Image-2.5-Flare cross the Bio High threshold or the Cyber High threshold. We, regardless and as done in GPT-Images 2.0, take a precautionary approach and apply mitigations appropriate for a model treated as High capability in the biological-risk domain.", "url": "https://deploymentsafety.openai.com/chatgpt-images-2-5/", "content_sha256": "4707244c63d42d903e09d7a0afb42e8f9c52e6fc2da07537e49d7263df057081", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-oss-20b", "source_id": "openai/gpt-oss-20b", "name": "GPT OSS 20B", "developer": "OpenAI", "url": "https://huggingface.co/openai/gpt-oss-20b", "description": "OpenAI's 21-billion-parameter open-weight reasoning model for low-latency, local, or specialized use, published as Apache-2.0 weights with an accompanying law-only usage statement.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "GPT OSS 20B", "description": "Open GPT reasoning model for self-hosted agents and controllable deployments", "family": "gpt-oss", "release_date": "2025-08-05", "last_updated": "2025-08-05", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-20b" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text and governs the published weights, tokenizer, and chat template. A separate USAGE_POLICY file asks users to comply with applicable law but adds no use-based restrictions beyond the license, so the release is open source.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "openai/gpt-oss-20b model weights, tokenizer, chat template, and original and Metal checkpoint artifacts", "kind": "web_terms", "url": "https://huggingface.co/openai/gpt-oss-20b/blob/6cee5e81ee83917806bbde320786a8fb61efebee/LICENSE", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the openai/gpt-oss-20b weight release at Hugging Face commit 6cee5e81ee83917806bbde320786a8fb61efebee, last modified 2025-08-26. The gpt-oss-120b sibling, downstream quantizations and fine-tunes, and independent hosting providers remain separate boundaries; OpenAI serves no managed API for this model.", "strengths": [ "OpenAI publishes Safetensors, original checkpoint, and Metal artifacts in one ungated repository with a repository license file", "The 20B sizing runs within 16 GB of memory with documented Transformers, vLLM, SGLang, Ollama, LM Studio, and Docker paths, and the smaller model can be fine-tuned on consumer hardware" ], "tradeoffs": [ "Neither the model card nor OpenAI's docs page carries dated release notes or a deprecation posture, so version identity rests on repository commits", "OpenAI's docs page is reference-only with zero rate limits, so every deployment is self-operated or third-party" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 8.08 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/openai/gpt-oss-20b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI docs page for gpt-oss-20b", "url": "https://developers.openai.com/api/docs/models/gpt-oss-20b", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-oss-20b.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://cdn.openai.com/pdf/419b6906-9da6-406c-a19d-1bb078ac7637/oai_gpt-oss_model_card.pdf", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-oss-safeguard-120b", "source_id": "openai/gpt-oss-safeguard-120b", "name": "GPT OSS Safeguard 120B", "developer": "OpenAI", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b", "description": "An ungated 117B safety-reasoning model built on the gpt-oss family that classifies text against caller-provided safety policies, released under Apache-2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "GPT OSS Safeguard 120B", "description": "Safety model for policy screening, moderation, and risk-aware routing workflows", "family": "gpt-oss", "release_date": "2025-10-29", "last_updated": "2025-10-29", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships a verbatim Apache License 2.0 file beside a USAGE_POLICY that adds no use restriction, and the card and announcement confirm the permissive grant for commercial deployment.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b/blob/main/LICENSE", "verified_at": "2026-09-18", "scope": "Official openai/gpt-oss-safeguard-120b weights, tokenizer, and bundled model code" } ], "access_boundary": "This record represents the openai/gpt-oss-safeguard-120b release of October 2025. The gpt-oss-120b base, the 20B safeguard sibling, downstream fine-tunes, and the OpenAI hosted services remain separate boundaries.", "strengths": [ "Ungated weights with the verbatim Apache-2.0 license file shipped in the repository", "The card documents the safety-policy classification usage and the technical report is published" ], "tradeoffs": [ "The MXFP4-quantized artifact set ships without full-precision or original checkpoints", "No first-party API or developers.openai.com model page exists for the release" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 7.36 }, "evidence": [ { "kind": "web", "label": "Official OpenAI weights repository and model card", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GPT OSS Safeguard announcement", "url": "https://openai.com/index/introducing-gpt-oss-safeguard/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-oss-safeguard-120b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://cdn.openai.com/pdf/08b7dee4-8bc6-4955-a219-7793fb69090c/Technical_report__Research_Preview_of_gpt_oss_safeguard.pdf", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-oss-safeguard-20b", "source_id": "openai/gpt-oss-safeguard-20b", "name": "GPT OSS Safeguard 20B", "developer": "OpenAI", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b", "description": "An ungated 21B safety-reasoning model built on the gpt-oss family that classifies text against caller-provided safety policies, released under Apache-2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "GPT OSS Safeguard 20B", "description": "Safety model for policy screening, moderation, and risk-aware routing workflows", "family": "gpt-oss", "release_date": "2025-10-29", "last_updated": "2025-10-29", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships a verbatim Apache License 2.0 file beside a USAGE_POLICY that adds no use restriction, and the card and announcement confirm the permissive grant for commercial deployment.", "license_evidence": [ { "license_id": "Apache-2.0", "kind": "web_terms", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/blob/main/LICENSE", "verified_at": "2026-09-18", "scope": "Official openai/gpt-oss-safeguard-20b weights, tokenizer, and bundled model code" } ], "access_boundary": "This record represents the openai/gpt-oss-safeguard-20b release of October 2025. The gpt-oss-20b base, the 120B safeguard sibling, downstream fine-tunes, and the OpenAI hosted services remain separate boundaries.", "strengths": [ "Ungated weights with the verbatim Apache-2.0 license file shipped in the repository", "The card documents the safety-policy classification usage and the technical report is published" ], "tradeoffs": [ "The MXFP4-quantized artifact set ships without full-precision or original checkpoints", "No first-party API or developers.openai.com model page exists for the release" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 7.36 }, "evidence": [ { "kind": "web", "label": "Official OpenAI weights repository and model card", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GPT OSS Safeguard announcement", "url": "https://openai.com/index/introducing-gpt-oss-safeguard/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-oss-safeguard-20b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://cdn.openai.com/pdf/08b7dee4-8bc6-4955-a219-7793fb69090c/Technical_report__Research_Preview_of_gpt_oss_safeguard.pdf", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-3-5-turbo", "source_id": "openai/gpt-3.5-turbo", "name": "GPT-3.5 Turbo", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-3.5-turbo", "description": "A legacy GPT model line served through OpenAI's API under the OpenAI Services Agreement, text-only, without downloadable model weights; OpenAI names GPT-5.6 Terra as its recommended replacement.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-3.5-turbo", "description": "Compact GPT model for low-latency assistance and high-volume workloads", "family": "gpt", "release_date": "2023-03-01", "last_updated": "2023-11-06", "knowledge_cutoff": "2021-09-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": false, "temperature": true }, "limits": { "context": 16385, "input": null, "output": 4096 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the GPT-3.5 Turbo release identified by the dated snapshot gpt-3.5-turbo-0125, announced March 2023. OpenAI announced its deprecation and documents the shutdown date of October 23, 2026 with GPT-5.6 Terra as the recommended replacement. The later GPT-4o and GPT-5 family lines, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "Chat Completions, Responses API, and fine-tuning are documented endpoints" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access ends at the documented shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official GPT-3.5 Turbo model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-3.5-turbo", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-3.5-turbo.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-3.5-turbo", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-4", "source_id": "openai/gpt-4", "name": "GPT-4", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-4", "description": "The flagship GPT-4 API line served under the OpenAI Services Agreement, text-only, without downloadable model weights; OpenAI names GPT-5.6 Sol as its recommended replacement.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-4", "description": "GPT model for general reasoning, writing, coding, and tool-assisted tasks", "family": "gpt", "release_date": "2023-11-06", "last_updated": "2024-04-09", "knowledge_cutoff": "2023-11", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 8192, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the GPT-4 release identified by the dated snapshot gpt-4-0613, announced November 2023. OpenAI announced its deprecation and documents the shutdown date of October 23, 2026 with GPT-5.6 Sol as the recommended replacement. The later GPT-4o, GPT-4.1, and GPT-5 family lines, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "Chat Completions, Responses API, Assistants, batch, and fine-tuning are documented endpoints" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access ends at the documented shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official GPT-4 model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-4.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-4", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-4-turbo", "source_id": "openai/gpt-4-turbo", "name": "GPT-4 Turbo", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-4-turbo", "description": "A vision-capable GPT-4 API line served under the OpenAI Services Agreement, without downloadable model weights; OpenAI names GPT-5.6 Sol as its recommended replacement.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-4 Turbo", "description": "Compact GPT model for low-latency assistance and high-volume workloads", "family": "gpt", "release_date": "2023-11-06", "last_updated": "2024-04-09", "knowledge_cutoff": "2023-12", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the GPT-4 Turbo release identified by the dated snapshot gpt-4-turbo-2024-04-09, announced November 2023. OpenAI announced its deprecation and documents the shutdown date of October 23, 2026 with GPT-5.6 Sol as the recommended replacement. The later GPT-4o, GPT-4.1, and GPT-5 family lines, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "A 128,000-token context window with image input is documented" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access ends at the documented shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official GPT-4 Turbo model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4-turbo", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-4-turbo.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-4-turbo", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-4-1", "source_id": "openai/gpt-4.1", "name": "GPT-4.1", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-4.1", "description": "A text-and-image model release exposed as a stable dated snapshot and alias through OpenAI's managed developer API, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-4.1", "description": "Long-lived GPT workhorse for coding, instruction following, and production apps", "family": "gpt", "release_date": "2025-04-14", "last_updated": "2025-04-14", "knowledge_cutoff": "2024-04", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1047576, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-4.1 as an API service governed by the OpenAI Services Agreement and incorporated service-specific policies. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-4.1 model through OpenAI API Services and incorporated service-specific terms", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-04" } ], "access_boundary": "This record represents the GPT-4.1 release whose API page identifies gpt-4.1-2025-04-14 as the fixed snapshot behind the gpt-4.1 alias. The OpenAI API platform, ChatGPT, fine-tuned derivatives, and later GPT releases are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on an alias", "The first-party model page records supported modalities, endpoints, features, and limits in one release-specific source" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed API rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.66 }, "evidence": [ { "kind": "web", "label": "Official GPT-4.1 model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4.1", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/ef71a4865a5821d8cd24f171a0100156a02e808e/models/openai/gpt-4.1.toml", "verified_at": "2026-09-04" } ], "metadata_verified_at": "2026-09-04", "verified_at": "2026-09-04", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-4.1", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-4-1-mini", "source_id": "openai/gpt-4.1-mini", "name": "GPT-4.1 mini", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-mini", "description": "The smaller, faster GPT-4.1 variant for instruction following and tool calling with a 1M-token context window and no reasoning step, exposed as the gpt-4.1-mini alias over the gpt-4.1-mini-2025-04-14 snapshot through OpenAI's managed developer API, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-4.1 mini", "description": "Affordable GPT-4.1 lane for fast coding help and structured extraction", "family": "gpt-mini", "release_date": "2025-04-14", "last_updated": "2025-04-14", "knowledge_cutoff": "2024-04", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1047576, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-4.1 Mini as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-4.1 Mini model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-4.1 Mini release whose API page identifies gpt-4.1-mini-2025-04-14 as the fixed snapshot behind the gpt-4.1-mini alias. OpenAI's February 2026 retirement of GPT-4.1 mini from ChatGPT explicitly leaves the API unchanged, and the API deprecation page carries no entry for this snapshot as of review. ChatGPT, the GPT-4.1 and GPT-4.1 nano siblings, later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "The managed API supports fine-tuning on the mini snapshot alongside chat, responses, assistants, and batch endpoints" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "OpenAI's own page steers new complex work toward GPT-5 Mini, and the ChatGPT retirement scopes this line's future to the API" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.66 }, "evidence": [ { "kind": "web", "label": "Official GPT-4.1 Mini model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-mini", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "OpenAI announcement scoping the GPT-4.1 mini retirement to ChatGPT with no API changes", "url": "https://openai.com/index/retiring-gpt-4o-and-older-models/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-4.1-mini.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-mini", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-4-1-nano", "source_id": "openai/gpt-4.1-nano", "name": "GPT-4.1 nano", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-nano", "description": "The fastest, cheapest GPT-4.1 variant for instruction following and tool calling with a 1M-token context window and no reasoning step, exposed as the gpt-4.1-nano alias over the gpt-4.1-nano-2025-04-14 snapshot through OpenAI's managed developer API, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-4.1 nano", "description": "Tiny GPT-4.1 option for classification, routing, and very high-volume tasks", "family": "gpt-nano", "release_date": "2025-04-14", "last_updated": "2025-04-14", "knowledge_cutoff": "2024-04", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1047576, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-4.1 Nano as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-4.1 Nano model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-4.1 Nano release whose API page identifies gpt-4.1-nano-2025-04-14 as the fixed snapshot behind the gpt-4.1-nano alias; OpenAI's deprecation page lists that snapshot for shutdown on 2026-10-23 with gpt-5.6-luna as the substitute model. ChatGPT, the GPT-4.1 and GPT-4.1 Mini siblings, later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "The deprecation page records an announcement date, shutdown date, and substitute model for the snapshot" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The only documented snapshot shuts down on 2026-10-23, about five weeks after review, with migration directed at gpt-5.6-luna" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.66 }, "evidence": [ { "kind": "web", "label": "Official GPT-4.1 Nano model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-nano", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-4.1-nano.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-nano", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-4o", "source_id": "openai/gpt-4o", "name": "GPT-4o", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-4o", "description": "A text-and-image model release exposed as the gpt-4o alias over three dated snapshots through OpenAI's managed developer API, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-4o", "description": "Omni-era GPT for multimodal chat, practical coding, and general assistants", "family": "gpt", "release_date": "2024-05-13", "last_updated": "2024-08-06", "knowledge_cutoff": "2023-09", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 16384 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-4o as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-4o model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the GPT-4o release whose API page lists gpt-4o-2024-08-06 as the default snapshot behind the gpt-4o alias alongside the gpt-4o-2024-05-13 and gpt-4o-2024-11-20 snapshots; OpenAI's deprecation page lists gpt-4o-2024-05-13 for shutdown on 2026-10-23. ChatGPT, gpt-4o-mini, audio and realtime variants, fine-tuned derivatives, and later GPT releases are separate boundaries.", "strengths": [ "OpenAI documents three dated snapshots and the alias default, so applications can pin a specific revision", "The deprecation page records announcement and shutdown dates per snapshot rather than per alias" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The earliest snapshot is already scheduled for shutdown, and movement of the alias between snapshots is controlled by the managed service" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.66 }, "evidence": [ { "kind": "web", "label": "Official GPT-4o model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4o", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/openai/gpt-4o.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "medium risk", "domains": [ "cyber", "bio_chem", "autonomy" ], "determination": "below_threshold", "scope": "weights", "statement": "After reviewing the results from the Preparedness evaluations, the Safety Advisory Group recommended classifying GPT‑4o before mitigations as borderline medium risk for persuasion, and low risk in all others.", "url": "https://openai.com/index/gpt-4o-system-card/", "content_sha256": "3e61d3a34570bc2f3e9d76e68bb708347f3096c4b3fc925983040fa34cb35be6", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-4o-mini", "source_id": "openai/gpt-4o-mini", "name": "GPT-4o mini", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-4o-mini", "description": "The fast, affordable GPT-4o variant for focused tasks and fine-tuning, exposed as the gpt-4o-mini alias over the gpt-4o-mini-2024-07-18 snapshot through OpenAI's managed developer API, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-4o mini", "description": "Small omni GPT for cheap multimodal assistance and production-scale traffic", "family": "gpt-mini", "release_date": "2024-07-18", "last_updated": "2024-07-18", "knowledge_cutoff": "2023-09", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 16384 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-4o Mini as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-4o Mini model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-4o Mini release whose API page identifies gpt-4o-mini-2024-07-18 as the fixed snapshot behind the gpt-4o-mini alias; the API deprecation page carries no entry for this snapshot as of review. ChatGPT, GPT-4o, the transcribe, realtime, audio, and search mini variants, fine-tuned derivatives, and later GPT releases are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "The managed API supports fine-tuning on the mini snapshot, and larger-model outputs can be distilled into it for lower cost and latency" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed API rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.66 }, "evidence": [ { "kind": "web", "label": "Official GPT-4o Mini model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4o-mini", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-4o-mini.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-4o-mini", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5", "source_id": "openai/gpt-5", "name": "GPT-5", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5", "description": "A text-and-image model release exposed as the gpt-5 alias over a dated snapshot through OpenAI's managed developer API, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5", "description": "Original GPT-5 workhorse for reasoning, coding, writing, and tool workflows", "family": "gpt", "release_date": "2025-08-07", "last_updated": "2025-08-07", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5 as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5 model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the GPT-5 release whose API page identifies gpt-5-2025-08-07 as the fixed snapshot behind the gpt-5 alias; OpenAI's deprecation page lists that snapshot for shutdown on 2026-12-11. ChatGPT, the gpt-5-mini and gpt-5-nano variants, later GPT-5.x releases, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "The deprecation page records an announcement date, shutdown date, and recommended replacement for the snapshot" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The only documented snapshot is already scheduled for shutdown, and no independent cloud path is named on the first-party page" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.66 }, "evidence": [ { "kind": "web", "label": "Official GPT-5 model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/openai/gpt-5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High capability", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Similarly to ChatGPT agent, we have decided to treat gpt-5-thinking as High capability in the Biological and Chemical domain under our Preparedness Framework, activating the associated safeguards. While we do not have definitive evidence that this model could meaningfully help a novice to create severe biological harm–our defined threshold for High capability–we have chosen to take a precautionary approach.", "url": "https://deploymentsafety.openai.com/gpt-5", "content_sha256": "e574578cbb11f0c71ea9513498af1474fc9b5718b9465ed0f89bf1a88bc62e76", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-mini", "source_id": "openai/gpt-5-mini", "name": "GPT-5 Mini", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5-mini", "description": "The cost-efficient GPT-5 variant for low-latency, high-volume work, exposed as the gpt-5-mini alias over the gpt-5-mini-2025-08-07 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5 Mini", "description": "Small GPT-5 for responsive agents, coding help, and everyday automation", "family": "gpt-mini", "release_date": "2025-08-07", "last_updated": "2025-08-07", "knowledge_cutoff": "2024-05-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5 Mini as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5 Mini model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5 Mini release whose API page identifies gpt-5-mini-2025-08-07 as the fixed snapshot behind the gpt-5-mini alias; OpenAI's deprecation page lists that snapshot for shutdown on 2026-12-11 with gpt-5.6-terra as the recommended replacement. ChatGPT, the GPT-5, GPT-5 nano, and later GPT-5.x releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API and Azure Foundry, whose reasoning-models page names gpt-5-mini with dated feature support" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The only documented snapshot is already scheduled for shutdown on 2026-12-11, with migration directed at gpt-5.6-terra" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5 Mini model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-mini", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5-mini", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5-mini.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/gpt-5", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-nano", "source_id": "openai/gpt-5-nano", "name": "GPT-5 Nano", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5-nano", "description": "The fastest, cheapest GPT-5 variant for summarization and classification, exposed as the gpt-5-nano alias over the gpt-5-nano-2025-08-07 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5 Nano", "description": "Tiny GPT-5 lane for routing, extraction, classification, and bulk jobs", "family": "gpt-nano", "release_date": "2025-08-07", "last_updated": "2025-08-07", "knowledge_cutoff": "2024-05-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5 Nano as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5 Nano model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5 Nano release whose API page identifies gpt-5-nano-2025-08-07 as the fixed snapshot behind the gpt-5-nano alias; OpenAI's deprecation page lists that snapshot for shutdown on 2026-12-11 with gpt-5.6-luna as the recommended replacement. ChatGPT, the GPT-5, GPT-5 Mini, and later GPT-5.x releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API and Azure Foundry, whose reasoning-models page names gpt-5-nano with dated feature support" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The only documented snapshot is already scheduled for shutdown on 2026-12-11, with migration directed at gpt-5.6-luna" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5 Nano model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-nano", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5-nano", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5-nano.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/gpt-5", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-pro", "source_id": "openai/gpt-5-pro", "name": "GPT-5 Pro", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5-pro", "description": "The precision GPT-5 tier with more test-time compute for demanding reasoning tasks, exposed as the gpt-5-pro alias over the gpt-5-pro-2025-10-06 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5 Pro", "description": "Higher-accuracy GPT-5 tier for tough analysis, coding reviews, and planning", "family": "gpt-pro", "release_date": "2025-10-06", "last_updated": "2025-10-06", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 272000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5 Pro as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5 Pro model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5 Pro release whose API page identifies gpt-5-pro-2025-10-06 as the fixed snapshot behind the gpt-5-pro alias; OpenAI's deprecation page lists that snapshot for shutdown on 2026-12-11 with gpt-5.6-sol in pro reasoning mode as the recommended replacement. ChatGPT, the GPT-5, GPT-5 Mini, GPT-5 Nano, and later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API and Azure Foundry, whose reasoning-models page names gpt-5-pro with dated feature support" ], "tradeoffs": [ "Serving is narrower than the base model: Responses API only, with fixed high reasoning effort and no code interpreter", "The only documented snapshot is already scheduled for shutdown on 2026-12-11, with migration directed at gpt-5.6-sol in pro reasoning mode" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5 Pro model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-pro", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5-pro", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5-pro.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/gpt-5", "verified_at": "2026-10-09", "research_confidence": "low" } ] }, { "id": "model-openai-gpt-5-codex", "source_id": "openai/gpt-5-codex", "name": "GPT-5-Codex", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5-codex", "description": "The September 2025 GPT-5 agentic-coding release for Codex, served Responses-only as the rotating gpt-5-codex alias with no fixed dated snapshot and retired on July 23, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5-Codex", "description": "Coding-optimized GPT model for repository edits, reviews, and agentic software work", "family": "gpt-codex", "release_date": "2025-09-15", "last_updated": "2025-09-15", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5-Codex as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5-Codex model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5-Codex line launched September 15, 2025 and served only as the rotating gpt-5-codex alias, whose underlying snapshot was regularly updated with no fixed dated revision; OpenAI shut the API down on July 23, 2026, naming gpt-5.6-sol the substitute. Codex CLI, IDE, cloud, and review application behaviors, later Codex generations, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "The September 2025 launch announcement pins the release and its Codex CLI, IDE, cloud, and review surfaces", "The dedicated model page documents text and image input, the Responses-only endpoint set, and the reasoning cutoff" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The alias rotated with no pinnable snapshot, and the line is retired with migration directed at gpt-5.6-sol" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official GPT-5-Codex model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-codex", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5-Codex launch announcement (September 15, 2025)", "url": "https://openai.com/index/introducing-upgrades-to-codex/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5-codex.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High risk", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Like GPT-5, we are treating GPT-5-Codex as High risk in the Biological and Chemical domain.", "url": "https://deploymentsafety.openai.com/gpt-5-codex", "content_sha256": "e67de04524efe37e6653c19b8719a3d2e81a1a73a08dfad9aa8e4e99da0d02c1", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-1", "source_id": "openai/gpt-5.1", "name": "GPT-5.1", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.1", "description": "The GPT-5.1 flagship line for coding and agentic work with configurable reasoning effort, exposed as the gpt-5.1 alias over the gpt-5.1-2025-11-13 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.1", "description": "Sharper GPT-5 generation for coding, product work, and tool-assisted tasks", "family": "gpt", "release_date": "2025-11-13", "last_updated": "2025-11-13", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.1 as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.1 model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.1 release whose API page identifies gpt-5.1-2025-11-13 as the fixed snapshot behind the gpt-5.1 alias; the API deprecation page carries no retirement entry for this base line as of review, while its chat-latest and codex variant endpoints shut down in July and August 2026. ChatGPT, the retired 5.1 chat and codex variants, the GPT-5.1 Mini/Nano siblings, later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API with chat, responses, and batch endpoints, and Azure Foundry, whose reasoning-models page names gpt-5.1 with dated feature support" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The generation's chat-latest and codex variant endpoints already shut down in July and August 2026, leaving the base API line as the surviving 5.1 surface" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.1 model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.1", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.1.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High risk", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "As we did for GPT-5 at launch, we are continuing to treat GPT-5.1 as High risk in the Biological and Chemical domain, and continuing to apply the corresponding safeguards.", "url": "https://deploymentsafety.openai.com/gpt-5-1", "content_sha256": "93d130bcf1ae05d14a7dcaa9d202a724caa7d4f20f409cc909163f2a824ddaf1", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-1-codex", "source_id": "openai/gpt-5.1-codex", "name": "GPT-5.1 Codex", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex", "description": "The November 2025 GPT-5.1 agentic-coding release for long-running coding tasks, served Responses-only as the rotating gpt-5.1-codex alias with no fixed dated snapshot and retired on July 23, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5.1 Codex", "description": "Codex GPT for repository edits, code review, and practical software agents", "family": "gpt-codex", "release_date": "2025-11-13", "last_updated": "2025-11-13", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.1-Codex as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.1-Codex model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.1-Codex line launched November 13, 2025 and served only as the rotating gpt-5.1-codex alias, whose underlying snapshot was regularly updated with no fixed dated revision; OpenAI shut the API down on July 23, 2026, naming gpt-5.6-sol the substitute. The Max line that replaced it as the Codex default, the Mini sibling, Codex application behaviors, later Codex generations, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "The November 2025 launch announcement pins the API release alongside the Mini sibling for long-running agentic coding tasks", "The dedicated model page documents text and image input, the Responses-only endpoint set, and the reasoning cutoff" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The alias rotated with no pinnable snapshot, and the line is retired with migration directed at gpt-5.6-sol" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.1-Codex model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.1 launch announcement (November 13, 2025)", "url": "https://openai.com/index/gpt-5-1-for-developers/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.1-codex.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-1-codex-max", "source_id": "openai/gpt-5.1-codex-max", "name": "GPT-5.1 Codex Max", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex-max", "description": "The November 2025 frontier agentic-coding release for long-running tasks, served Responses-only as the rotating gpt-5.1-codex-max alias with no fixed dated snapshot and retired on July 23, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5.1 Codex Max", "description": "Coding-optimized GPT model for repository edits, reviews, and agentic software work", "family": "gpt-codex", "release_date": "2025-11-13", "last_updated": "2025-11-13", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.1-Codex-Max as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.1-Codex-Max model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.1-Codex-Max line launched November 19, 2025, which replaced GPT-5.1-Codex as the default model in Codex surfaces and was served only as the rotating gpt-5.1-codex-max alias with no fixed dated revision; OpenAI shut the API down on July 23, 2026, naming gpt-5.6-sol the substitute. The Codex and Mini siblings, Codex application behaviors, later Codex generations, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "The November 2025 launch announcement pins the release and its succession of GPT-5.1-Codex as the Codex default", "The dedicated model page documents text and image input, the Responses-only endpoint set, and the reasoning cutoff" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The alias rotated with no pinnable snapshot, and the line is retired with migration directed at gpt-5.6-sol" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.1-Codex-Max model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex-max", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.1-Codex-Max launch announcement (November 19, 2025)", "url": "https://openai.com/index/gpt-5-1-codex-max/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.1-codex-max.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High risk", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "We are treating GPT-5.1-Codex-Max as High risk in the Biological and Chemical domain.", "url": "https://deploymentsafety.openai.com/gpt-5-1-codex-max", "content_sha256": "b8ac2c39fdabfae17d453a71296d7d3ae659f40de1415cb030377cf859fcb548", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-1-codex-mini", "source_id": "openai/gpt-5.1-codex-mini", "name": "GPT-5.1 Codex mini", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex-mini", "description": "The smaller November 2025 GPT-5.1 agentic-coding release, served Responses-only as the rotating gpt-5.1-codex-mini alias with no fixed dated snapshot and retired on July 23, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5.1 Codex mini", "description": "Coding-optimized GPT model for repository edits, reviews, and agentic software work", "family": "gpt-codex", "release_date": "2025-11-13", "last_updated": "2025-11-13", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.1-Codex Mini as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.1-Codex Mini model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.1-Codex Mini line launched November 13, 2025 as the smaller sibling of GPT-5.1-Codex, served only as the rotating gpt-5.1-codex-mini alias with no fixed dated revision; OpenAI shut the API down on July 23, 2026, naming gpt-5.6-terra the substitute. The Codex and Max siblings, Codex application behaviors, later Codex generations, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "The November 2025 launch announcement pins the API release as the smaller GPT-5.1-Codex sibling", "The dedicated model page documents text and image input, the Responses-only endpoint set, and evals and stored-completion support" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The alias rotated with no pinnable snapshot, and the line is retired with migration directed at gpt-5.6-terra" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.1-Codex Mini model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex-mini", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.1 launch announcement (November 13, 2025)", "url": "https://openai.com/index/gpt-5-1-for-developers/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.1-codex-mini.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex-mini", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-2", "source_id": "openai/gpt-5.2", "name": "GPT-5.2", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.2", "description": "The previous GPT-5 flagship generation for professional work with configurable reasoning effort, exposed as the gpt-5.2 alias over the gpt-5.2-2025-12-11 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.2", "description": "Reliable GPT generation for broad coding, writing, and tool-assisted product work", "family": "gpt", "release_date": "2025-12-11", "last_updated": "2025-12-11", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.2 as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.2 model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.2 release whose API page identifies gpt-5.2-2025-12-11 as the fixed snapshot behind the gpt-5.2 alias; the API deprecation page carries no retirement entry for this base line as of review, while its chat-latest and codex variant endpoints shut down in July and August 2026. ChatGPT, the retired 5.2 chat and codex variants, later GPT releases including GPT-6 Astra, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API with chat, responses, and batch endpoints, and Azure Foundry, whose reasoning-models page names gpt-5.2" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "OpenAI's own page positions GPT-5.2 as the previous flagship and recommends GPT-6 Astra for new work, while the generation's chat and codex variants already shut down" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.2 model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.2", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.2", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.2.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High capability", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "We are treating this launch as High capability in the Biological and Chemical domain, activating the associated Preparedness safeguards. We do not have definitive evidence that these models could meaningfully help a novice to create severe biological harm, our defined threshold for High capability, and these models remain on the cusp of being able to reach this capability.", "url": "https://deploymentsafety.openai.com/gpt-5-2", "content_sha256": "f5c80ab6cc54aaac1dabbdc5a9f67d8c20678c027420f315e848ff8cfac876b1", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-2-codex", "source_id": "openai/gpt-5.2-codex", "name": "GPT-5.2 Codex", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.2-codex", "description": "The December 2025 agentic-coding release with configurable reasoning effort, served Responses-only as the rotating gpt-5.2-codex alias with no fixed dated snapshot and retired on July 23, 2026, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5.2 Codex", "description": "Code-specialist GPT for repository edits, reviews, and long-running software agents", "family": "gpt-codex", "release_date": "2025-12-11", "last_updated": "2025-12-11", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.2-Codex as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.2-Codex model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.2-Codex line launched December 18, 2025 and served only as the rotating gpt-5.2-codex alias with no fixed dated revision; OpenAI shut the API down on July 23, 2026, naming gpt-5.6-sol the substitute. Codex application behaviors, the GPT-5.3-Codex successor, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "The December 2025 launch announcement pins the release across Codex surfaces with API access to follow", "The dedicated model page documents configurable reasoning effort with hosted-shell and skills tools" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "The alias rotated with no pinnable snapshot, and the line is retired with migration directed at gpt-5.6-sol" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.2-Codex model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.2-codex", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.2-Codex launch announcement (December 18, 2025)", "url": "https://openai.com/index/gpt-5-2-codex/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.2-codex.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "As we did for other models in the GPT-5 series, we are treating GPT-5.2-Codex as High risk in the Biological and Chemical domain, and continuing to apply the corresponding safeguards.", "url": "https://deploymentsafety.openai.com/gpt-5-2-codex", "content_sha256": "aa89d91c3671bfd5b892e56d0fd43f50f4a2f88cf0cc7482653ec6c94a6b6851", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-2-pro", "source_id": "openai/gpt-5.2-pro", "name": "GPT-5.2 Pro", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.2-pro", "description": "An API-served GPT release pinned to the dated snapshot gpt-5.2-pro-2025-12-11, available through OpenAI's Responses API under the OpenAI Services Agreement, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5.2 Pro", "description": "Higher-accuracy GPT-5.2 variant for tougher reasoning and review workflows", "family": "gpt-pro", "release_date": "2025-12-11", "last_updated": "2025-12-11", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": false }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.2 Pro as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-18", "scope": "Commercial access to GPT-5.2 Pro through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the GPT-5.2 Pro release identified by the dated snapshot gpt-5.2-pro-2025-12-11, launched December 2025 and superseded by the GPT-5.5 Pro tier. The GPT-5.5 Pro successor, the standard GPT-5.2 line, the OpenAI API platform, ChatGPT, and later GPT releases remain separate service or product boundaries.", "strengths": [ "OpenAI documents a pinnable dated snapshot with launch date, modalities, and endpoints", "The model page states the superseded tier explicitly and names the recommended successor" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The release is served only through the Responses API with no documented third-party path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.66 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.2 Pro model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.2-pro", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GPT-5.2 announcement", "url": "https://openai.com/index/introducing-gpt-5-2/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.2-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/gpt-5-2", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-3-codex", "source_id": "openai/gpt-5.3-codex", "name": "GPT-5.3 Codex", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.3-codex", "description": "The February 2026 agentic-coding release for Codex, served Responses-only as the gpt-5.3-codex alias with no fixed dated snapshot and no API retirement entry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5.3 Codex", "description": "Coding-optimized GPT model for repository edits, reviews, and agentic software work", "family": "gpt-codex", "release_date": "2026-02-05", "last_updated": "2026-02-05", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.3-Codex as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.3-Codex model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.3-Codex line launched February 5, 2026 and served as the gpt-5.3-codex alias with no fixed dated revision, so applications cannot pin a dated snapshot; the API deprecation page carries no retirement entry for this line as of review. The Spark research preview, Codex application behaviors, earlier Codex generations, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "The February 2026 launch announcement pins the release and its advance over GPT-5.2-Codex and GPT-5.2", "The dedicated model page documents text and image input with hosted-shell and skills tools on the Responses endpoint" ], "tradeoffs": [ "No documented fixed dated snapshot exists, so the record cannot pin behavior the way dated-snapshot records can", "No developer-published weights are available for self-operation or independent deployment" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.3-Codex model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.3-codex", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.3-Codex launch announcement (February 5, 2026)", "url": "https://openai.com/index/introducing-gpt-5-3-codex/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official deprecation page checked: no gpt-5.3-codex retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.3-codex.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "cyber", "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "As discussed in more detail below, this is the first launch we are treating as High capability in Cybersecurity. Thus, we consider the safeguards associated with High cybersecurity capability to be a necessary precaution. We are also treating GPT-5.3-Codex as High risk in the Biological and Chemical domain, and applying the same safeguards as for our other deployments of models that reach this designation.", "url": "https://deploymentsafety.openai.com/gpt-5-3-codex", "content_sha256": "02b9fa09a776a73fd234a46cd53134cd0a69f5eda0a429ab1b0eb3d3391b5fd9", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-4", "source_id": "openai/gpt-5.4", "name": "GPT-5.4", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.4", "description": "The GPT-5.4 flagship line for professional work with a 1M-token context window, exposed as the gpt-5.4 alias over the gpt-5.4-2026-03-05 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.4", "description": "Agent-ready GPT for coding and computer-use workflows at a lower cost", "family": "gpt", "release_date": "2026-03-05", "last_updated": "2026-03-05", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.4 as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.4 model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.4 release whose API page identifies gpt-5.4-2026-03-05 as the fixed snapshot behind the gpt-5.4 alias; the API deprecation page carries no retirement entry for this base line as of review. ChatGPT, the GPT-5.4 Mini, Nano, Pro, and invite-only Cyber siblings, later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API with a 1M-token context window and computer-use tool support, and Azure Foundry, whose reasoning-models page names gpt-5.4" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed APIs rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.4 model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.4", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.4.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High capability", "domains": [ "bio_chem", "cyber" ], "determination": "precautionary", "scope": "weights", "statement": "As we did for gpt-5.2-thinking before it, we are continuing to treat gpt-5.4-thinking as High capability in the Biological and Chemical domain. We have applied the corresponding safeguards for this model as described in the GPT-5 system card. As we did for gpt-5.3-codex, we are treating gpt-5.4-thinking as High capability in the Cybersecurity domain, and applied safeguards as described in the Safeguard section below.", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking", "content_sha256": "4b35355dfe6f143b53c8d6127467a7c125b23468a5849707caf5eb744ae9657f", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-4-mini", "source_id": "openai/gpt-5.4-mini", "name": "GPT-5.4 mini", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-mini", "description": "The efficient GPT-5.4 variant for coding, computer use, and subagents at high volume, exposed as the gpt-5.4-mini alias over the gpt-5.4-mini-2026-03-17 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.4 mini", "description": "Strong small GPT for coding subagents, quick tool use, and high-volume work", "family": "gpt-mini", "release_date": "2026-03-17", "last_updated": "2026-03-17", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.4 Mini as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.4 Mini model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.4 Mini release whose API page identifies gpt-5.4-mini-2026-03-17 as the fixed snapshot behind the gpt-5.4-mini alias; the API deprecation page carries no retirement entry for this line as of review. ChatGPT, the GPT-5.4, GPT-5.4 Nano, and GPT-5.4 Pro siblings, later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API with chat, responses, and batch endpoints, and Azure Foundry, whose reasoning-models page names gpt-5.4-mini" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed APIs rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.4 Mini model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-mini", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.4-mini", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.4-mini.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High capability", "domains": [ "bio_chem", "cyber", "autonomy" ], "determination": "below_threshold", "scope": "weights", "statement": "Based on our Preparedness evaluations, we have determined that GPT-5.4 mini as below High capability across biochemical, cybersecurity, and AI-self improvement domains.", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking", "content_sha256": "4b35355dfe6f143b53c8d6127467a7c125b23468a5849707caf5eb744ae9657f", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-4-nano", "source_id": "openai/gpt-5.4-nano", "name": "GPT-5.4 nano", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-nano", "description": "The cheapest GPT-5.4 variant for simple high-volume work such as classification and extraction, exposed as the gpt-5.4-nano alias over the gpt-5.4-nano-2026-03-17 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.4 nano", "description": "Cheapest GPT-5.4 lane for simple routing, extraction, and bulk automation", "family": "gpt-nano", "release_date": "2026-03-17", "last_updated": "2026-03-17", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 400000, "input": 272000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.4 Nano as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.4 Nano model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.4 Nano release whose API page identifies gpt-5.4-nano-2026-03-17 as the fixed snapshot behind the gpt-5.4-nano alias; the API deprecation page carries no retirement entry for this line as of review. ChatGPT, the GPT-5.4, GPT-5.4 Mini, and GPT-5.4 Pro siblings, later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API with chat, responses, and batch endpoints, and Azure Foundry, whose reasoning-models page names gpt-5.4-nano" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed APIs rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.4 Nano model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-nano", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.4-nano", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.4-nano.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-4-pro", "source_id": "openai/gpt-5.4-pro", "name": "GPT-5.4 Pro", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-pro", "description": "The precision GPT-5.4 tier with more test-time compute for demanding professional reasoning, exposed as the gpt-5.4-pro alias over the gpt-5.4-pro-2026-03-05 snapshot through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.4 Pro", "description": "More exact GPT-5.4 tier for demanding professional reasoning and agent tasks", "family": "gpt-pro", "release_date": "2026-03-05", "last_updated": "2026-03-05", "knowledge_cutoff": "2025-08-31", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.4 Pro as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.4 Pro model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.4 Pro release whose API page identifies gpt-5.4-pro-2026-03-05 as the fixed snapshot behind the gpt-5.4-pro alias; the API deprecation page carries no retirement entry for this line as of review. ChatGPT, the GPT-5.4, GPT-5.4 Mini, and GPT-5.4 Nano siblings, later GPT releases, fine-tuned derivatives, and the OpenAI API platform itself are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot that applications can pin instead of relying only on the alias", "Two managed serving paths: the first-party developer API with a 1M-token context window, and Azure Foundry, whose reasoning-models page names gpt-5.4-pro" ], "tradeoffs": [ "Serving is narrower than the base model: Responses API only, with no Chat Completions and no batch endpoint", "No developer-published weights are available for self-operation or independent deployment" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.4 Pro model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-pro", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official OpenAI model deprecation schedule", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.4-pro", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.4-pro.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-5-5", "source_id": "openai/gpt-5.5", "name": "GPT-5.5", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.5", "description": "A text-and-image model release exposed as a dated snapshot and alias through OpenAI's managed developer API and Azure Foundry, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.5", "description": "Default frontier GPT for coding, computer use, research, and knowledge work", "family": "gpt", "release_date": "2026-04-23", "last_updated": "2026-04-23", "knowledge_cutoff": "2025-12-01", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.5 as an API service governed by the OpenAI Services Agreement and incorporated service-specific policies. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.5 model through OpenAI API Services and incorporated service-specific terms", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the GPT-5.5 release whose API page identifies gpt-5.5-2026-04-23 as the fixed default snapshot behind the gpt-5.5 alias. The OpenAI API platform, Azure Foundry hosting of the same snapshot, ChatGPT, GPT-5.5 Thinking, the separate gpt-5.5-pro and gpt-5.5-instant variants, fine-tuned derivatives, and later GPT releases are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot (gpt-5.5-2026-04-23) that applications can pin instead of relying only on an alias", "The first-party model page plus the Azure Foundry listing establish two independent managed serving paths with documented modalities, endpoints, and limits" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed APIs rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.5 model, snapshot, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official GPT-5.5 launch announcement", "url": "https://openai.com/index/introducing-gpt-5-5/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Azure Foundry models listing showing gpt-5.5 availability", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official deprecation page checked: no gpt-5.5 retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/openai/gpt-5.5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High capability", "domains": [ "bio_chem", "cyber" ], "determination": "precautionary", "scope": "weights", "statement": "As we did for GPT-5.4 Thinking before it, we are continuing to treat GPT-5.5 as High capability in the Biological and Chemical domain. We have applied the corresponding safeguards for this model as described in the GPT-5 system card. As we did for GPT-5.3-Codex and GPT-5.4-thinking, we are treating GPT-5.5 as High capability in the Cybersecurity domain, but below Critical.", "url": "https://deploymentsafety.openai.com/gpt-5-5", "content_sha256": "3bbb721e756b622ebf2e96f4883a65d187d75fb2b08178334621b30adf42e059", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-5-instant", "source_id": "openai/gpt-5.5-instant", "name": "GPT-5.5 Instant", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/chat-latest", "description": "The low-latency GPT-5.5 tier that became ChatGPT's default model, served through OpenAI's managed developer API only as the rotating chat-latest alias and as dated preview deployments on Azure Foundry, without downloadable model weights or a pinnable first-party snapshot.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.5 Instant", "description": "Compact GPT model for low-latency assistance and high-volume workloads", "family": null, "release_date": "2026-05-05", "last_updated": "2026-05-28", "knowledge_cutoff": "2025-12-01", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 400000, "input": 400000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.5 Instant as an API service governed by the OpenAI Services Agreement and incorporated service-specific policies. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.5 Instant model served via the chat-latest alias through OpenAI API Services and incorporated service-specific terms", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the GPT-5.5 Instant release of May 5, 2026, exposed on the first-party API only as the moving chat-latest alias with no documented fixed snapshot: the alias page lists chat-latest as both model ID and sole snapshot and states the underlying snapshot is regularly updated. Azure Foundry hosts the same line as dated gpt-chat-latest preview deployments naming GPT-5.5 Instant as the OpenAI-side equivalent. The base GPT-5.5 release, GPT-5.5 Pro, the invite-only Cyber preview, ChatGPT application behaviors, Codex, the GPT-5.3 Instant predecessor, fine-tuned derivatives, and later Instant generations are separate service, application, or model boundaries.", "strengths": [ "The launch announcement pins the release date and the API identity, and Azure documents dated gpt-chat-latest preview deployments mapped to GPT-5.5 Instant", "First-party alias page plus system card establish Chat Completions and Responses support with image input" ], "tradeoffs": [ "No developer-published weights and no pinnable first-party snapshot exist, so production behavior can drift as the alias rotates", "The only third-party hosting path is an Azure preview-lifecycle product, and the live alias page recommends GPT-6 Astra for production API use" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 4.14 }, "evidence": [ { "kind": "web", "label": "Official chat-latest alias page: the API identity of GPT-5.5 Instant", "url": "https://developers.openai.com/api/docs/models/chat-latest", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official GPT-5.5 Instant launch announcement (May 5, 2026)", "url": "https://openai.com/index/gpt-5-5-instant/", "verified_at": "2026-09-12" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Azure Foundry models listing: dated gpt-chat-latest preview deployments", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official deprecation page checked: no retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/openai/gpt-5.5-instant.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High Capability", "domains": [ "bio_chem", "cyber" ], "determination": "precautionary", "scope": "weights", "statement": "GPT-5.5 Instant is our first Instant model to be treated as High Capability in the Biological and Chemical domain, as well as in the Cybersecurity domain. As such, we have applied biological and chemical, and cybersecurity, safeguards to this deployment.", "url": "https://deploymentsafety.openai.com/gpt-5-5-instant", "content_sha256": "2bf8c3dc34ade91c869350ba8540800f16079ec5abceddb6cf206bb81077d849", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-5-pro", "source_id": "openai/gpt-5.5-pro", "name": "GPT-5.5 Pro", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.5-pro", "description": "The highest-accuracy GPT-5.5 tier, served as a dated snapshot and alias through OpenAI's managed developer API with extra test-time compute for precision-heavy reasoning and coding, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-5.5 Pro", "description": "Highest-accuracy GPT-5.5 tier for slower, precision-heavy reasoning and coding", "family": "gpt-pro", "release_date": "2026-04-23", "last_updated": "2026-04-23", "knowledge_cutoff": "2025-12-01", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.5 Pro as an API service governed by the OpenAI Services Agreement and incorporated service-specific policies. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.5 Pro model through OpenAI API Services and incorporated service-specific terms", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the GPT-5.5 Pro release whose API page identifies gpt-5.5-pro-2026-04-23 as the fixed default snapshot behind the gpt-5.5-pro alias, served via the Responses API. The base GPT-5.5 release, the GPT-5.5 Instant tier, the invite-only Cyber preview, ChatGPT and Codex applications, Azure Foundry (which lists no gpt-5.5-pro row), fine-tuned derivatives, and later GPT releases are separate service, application, or model boundaries.", "strengths": [ "OpenAI documents a dated snapshot (gpt-5.5-pro-2026-04-23) that applications can pin, plus exact modalities, knowledge cutoff, endpoints, and rate limits", "The launch announcement and system card place Pro in the API and ChatGPT Pro, Business, and Enterprise tiers with stated pricing and safeguards" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Serving is narrower than the base model: Responses API only, with no Chat Completions, no cached-input discount, and no Azure Foundry listing" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.52 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.5 Pro model, snapshot, modality, cutoff, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.5-pro", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official GPT-5.5 launch announcement", "url": "https://openai.com/index/introducing-gpt-5-5/", "verified_at": "2026-09-12" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official deprecation page checked: no gpt-5.5-pro retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/openai/gpt-5.5-pro.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/gpt-5-5", "verified_at": "2026-10-09", "research_confidence": "low" } ] }, { "id": "model-openai-gpt-5-6-luna", "source_id": "openai/gpt-5.6-luna", "name": "GPT-5.6 Luna", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-luna", "description": "The cost-efficient GPT-5.6 tier for high-volume work, exposed as the gpt-5.6-luna tier name over OpenAI's managed developer API and Azure Foundry with no documented fixed dated snapshot and no downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.6 Luna", "description": "Cost-efficient GPT-5.6 model for fast, high-volume workloads", "family": "gpt-luna", "release_date": "2026-07-09", "last_updated": "2026-07-09", "knowledge_cutoff": "2026-02-16", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.6 Luna as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.6 Luna model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.6 Luna tier launched July 9, 2026: the official model page lists both the default snapshot and the snapshot inventory as gpt-5.6-luna with no documented fixed dated snapshot, so applications cannot pin a dated revision; OpenAI defines the tiers as durable capabilities that advance on their own cadence. The Sol and Terra sibling tiers, ChatGPT and Codex application surfaces, fine-tuned derivatives, and later generations are separate service, application, or model boundaries.", "strengths": [ "The July 2026 launch announcement pins the release date and the cost-efficient-tier positioning, and Azure documents gpt-5.6-luna availability as a second managed path", "The first-party model page documents modalities, knowledge cutoff, endpoints, and rate limits from the developer" ], "tradeoffs": [ "No documented fixed dated snapshot exists, so the record cannot pin behavior the way dated-snapshot records can", "No developer-published weights are available for self-operation or independent deployment" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 4.68 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.6 Luna model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-luna", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.6 launch announcement (July 9, 2026)", "url": "https://openai.com/index/gpt-5-6/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.6-luna", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official deprecation page checked: no gpt-5.6-luna retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.6-luna.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "cyber", "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Under our Preparedness Framework, we are treating Sol, Terra and Luna as High capability in both Cybersecurity and Biological and Chemical risk.", "url": "https://deploymentsafety.openai.com/gpt-5-6/", "content_sha256": "dd387153d747da76982a6da6a8aeafa03d491aa8abe5d6f4bd25745d3a68bcfa", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-6-sol", "source_id": "openai/gpt-5.6-sol", "name": "GPT-5.6 Sol", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-sol", "description": "The flagship tier of the July 2026 GPT-5.6 generation for professional work, exposed as the gpt-5.6-sol tier name over OpenAI's managed developer API and Azure Foundry with no documented fixed dated snapshot and no downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.6 Sol", "description": "Frontier GPT-5.6 model for complex professional work, coding, and agentic workflows", "family": "gpt-sol", "release_date": "2026-07-09", "last_updated": "2026-07-09", "knowledge_cutoff": "2026-02-16", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.6 Sol as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.6 Sol model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.6 Sol tier launched July 9, 2026: the official model page lists both the default snapshot and the snapshot inventory as gpt-5.6-sol with no documented fixed dated snapshot, so applications cannot pin a dated revision; OpenAI defines Sol, Terra, and Luna as durable capability tiers that advance on their own cadence. The Terra and Luna sibling tiers, the gpt-5.6 umbrella alias routing to Sol, ChatGPT and Codex application surfaces including the Sol Pro chat selection, fine-tuned derivatives, and later generations are separate service, application, or model boundaries.", "strengths": [ "The July 2026 launch announcement pins the release date and the flagship tier positioning, and Azure documents gpt-5.6-sol availability as a second managed path", "The first-party model page documents modalities, knowledge cutoff, endpoints, and rate limits from the developer" ], "tradeoffs": [ "No documented fixed dated snapshot exists, so the record cannot pin behavior the way dated-snapshot records can", "No developer-published weights are available for self-operation or independent deployment" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 4.68 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.6 Sol model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-sol", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.6 launch announcement (July 9, 2026)", "url": "https://openai.com/index/gpt-5-6/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.6-sol", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official deprecation page checked: no gpt-5.6-sol retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.6-sol.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "cyber", "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Under our Preparedness Framework, we are treating Sol, Terra and Luna as High capability in both Cybersecurity and Biological and Chemical risk.", "url": "https://deploymentsafety.openai.com/gpt-5-6/", "content_sha256": "dd387153d747da76982a6da6a8aeafa03d491aa8abe5d6f4bd25745d3a68bcfa", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-5-6-terra", "source_id": "openai/gpt-5.6-terra", "name": "GPT-5.6 Terra", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-terra", "description": "The balanced GPT-5.6 tier for everyday work, exposed as the gpt-5.6-terra tier name over OpenAI's managed developer API and Azure Foundry with no documented fixed dated snapshot and no downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-5.6 Terra", "description": "Balanced GPT-5.6 model for capable, cost-efficient everyday work", "family": "gpt-terra", "release_date": "2026-07-09", "last_updated": "2026-07-09", "knowledge_cutoff": "2026-02-16", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-5.6 Terra as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-5.6 Terra model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" } ], "access_boundary": "This record represents the GPT-5.6 Terra tier launched July 9, 2026: the official model page lists both the default snapshot and the snapshot inventory as gpt-5.6-terra with no documented fixed dated snapshot, so applications cannot pin a dated revision; OpenAI defines the tiers as durable capabilities that advance on their own cadence. The Sol and Luna sibling tiers, ChatGPT and Codex application surfaces, fine-tuned derivatives, and later generations are separate service, application, or model boundaries.", "strengths": [ "The July 2026 launch announcement pins the release date and the balanced-tier positioning, and Azure documents gpt-5.6-terra availability as a second managed path", "The first-party model page documents modalities, knowledge cutoff, endpoints, and rate limits from the developer" ], "tradeoffs": [ "No documented fixed dated snapshot exists, so the record cannot pin behavior the way dated-snapshot records can", "No developer-published weights are available for self-operation or independent deployment" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 4.68 }, "evidence": [ { "kind": "web", "label": "Official GPT-5.6 Terra model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-terra", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official GPT-5.6 launch announcement (July 9, 2026)", "url": "https://openai.com/index/gpt-5-6/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Azure Foundry reasoning-models page naming gpt-5.6-terra", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Official deprecation page checked: no gpt-5.6-terra retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-16" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-5.6-terra.toml", "verified_at": "2026-09-16" } ], "metadata_verified_at": "2026-09-16", "verified_at": "2026-09-16", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "cyber", "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Under our Preparedness Framework, we are treating Sol, Terra and Luna as High capability in both Cybersecurity and Biological and Chemical risk.", "url": "https://deploymentsafety.openai.com/gpt-5-6/", "content_sha256": "dd387153d747da76982a6da6a8aeafa03d491aa8abe5d6f4bd25745d3a68bcfa", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-6-astra", "source_id": "openai/gpt-6-astra", "name": "GPT-6 Astra", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-6-astra", "description": "A text-and-image model release exposed under a moving alias through OpenAI's managed developer API with documented Azure Foundry hosting, without a fixed dated snapshot or downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-6 Astra", "description": "GPT-6 Astra is OpenAI's most capable model for complex reasoning, coding, computer use, research, and document creation.", "family": "gpt-astra", "release_date": "2026-09-04", "last_updated": "2026-09-04", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-6 Astra as an API service governed by the OpenAI Services Agreement and incorporated service-specific policies. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-6 Astra model through OpenAI API Services and incorporated service-specific terms", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the GPT-6 Astra release as exposed under the gpt-6-astra alias: the official model page lists both the default snapshot and the snapshot inventory as gpt-6-astra with no documented fixed dated snapshot, so applications cannot pin a dated revision. The OpenAI API platform, Azure Foundry hosting of the same alias, ChatGPT, the separate gpt-6-astra-fast variant, fine-tuned derivatives, and later GPT releases are separate service, application, or model boundaries.", "strengths": [ "The first-party model page, launch announcement, and safety overview document modalities, endpoints, and safeguards from the developer", "The OpenAI announcement plus the Azure Foundry listing establish managed serving beyond the first-party API" ], "tradeoffs": [ "No documented fixed dated snapshot exists, so the record cannot pin behavior the way dated-snapshot records can", "No developer-published weights are available for self-operation, and availability was still rolling out at verification time" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 4.68 }, "evidence": [ { "kind": "web", "label": "Official GPT-6 Astra model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-6-astra", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official GPT-6 Astra launch announcement", "url": "https://openai.com/index/gpt-6-astra/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Azure Foundry models listing showing gpt-6-astra availability", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official deprecation page checked: no gpt-6-astra retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/openai/gpt-6-astra.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "Critical", "domains": [ "cyber" ], "determination": "determined", "scope": "weights", "statement": "Today, we are releasing GPT-6 Astra, the most capable model we have ever broadly deployed. Astra is our first model to reach the Critical level of cybersecurity capability under our Preparedness Framework.", "url": "https://deploymentsafety.openai.com/gpt-6-astra/", "content_sha256": "8f13bca387cc4990574bb106556d3a678b7c17cc6c8df0aa4e83df36a4367a82", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-6-luna", "source_id": "openai/gpt-6-luna", "name": "GPT-6 Luna", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-6-luna", "description": "The efficient GPT-6 tier released September 22, 2026 for focused, high-volume tasks, exposed as the gpt-6-luna name over OpenAI's managed developer API and Azure Foundry with no documented fixed dated snapshot on OpenAI's page and no downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-6 Luna", "description": "OpenAI's most efficient model for focused, high-volume tasks", "family": "gpt-luna", "release_date": "2026-09-22", "last_updated": "2026-09-22", "knowledge_cutoff": "2026-05-18", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-6 Luna as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-6 Luna model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the GPT-6 Luna release of September 22, 2026 as exposed under the gpt-6-luna name: OpenAI's model page lists the default snapshot and the snapshot inventory as gpt-6-luna with no dated snapshot, so applications cannot pin a dated revision through OpenAI, although Azure Foundry documents the version as gpt-6-luna (2026-09-22). OpenAI changed the name in place on September 25, 2026 (an image-encoding fix). The GPT-5.6 Luna tier, GPT-6 Sol, GPT-6.1 Sol and GPT-6 Astra, ChatGPT and Codex surfaces, the Decisions API that OpenAI released with this model, fine-tuned derivatives, and later generations are separate service, application, or model boundaries.", "strengths": [ "OpenAI's changelog pins the September 22, 2026 release, the model page documents modalities, limits, endpoints, and rate limits, and the model catalog lists it among the current GPT-6 models", "The deprecations table names the model as the replacement for gpt-5.4-nano, and Azure Foundry documents a dated version with no access request" ], "tradeoffs": [ "OpenAI documents no fixed dated snapshot and changed the name's behavior in place on September 25, 2026, so the record cannot pin behavior the way dated-snapshot records can", "No developer-published weights are available for self-operation or independent deployment" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Official GPT-6 Luna model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-6-luna", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official API changelog: GPT-6 Sol and GPT-6 Luna release (Sep 22) and image-encoding fix (Sep 25)", "url": "https://developers.openai.com/api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Azure Foundry Models sold by Azure listing gpt-6-luna (2026-09-22)", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Azure OpenAI reasoning-models page: gpt-6-luna availability and version date", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official deprecations page: gpt-6-luna named as a replacement, no retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official model catalog listing GPT-6 Luna among current models", "url": "https://developers.openai.com/api/docs/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/openai/gpt-6-luna.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "cyber", "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Under our Preparedness Framework, we are treating GPT-6 Sol and GPT-6 Luna as High capability in both Cybersecurity and Biological and Chemical domains.", "url": "https://deploymentsafety.openai.com/gpt-6-astra/", "content_sha256": "8f13bca387cc4990574bb106556d3a678b7c17cc6c8df0aa4e83df36a4367a82", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-6-sol", "source_id": "openai/gpt-6-sol", "name": "GPT-6 Sol", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-6-sol", "description": "The GPT-6 Sol release of September 22, 2026 for complex coding and agentic workflows, exposed as the gpt-6-sol name over OpenAI's managed developer API and Azure Foundry with no documented fixed dated snapshot on OpenAI's page and no downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-6 Sol", "description": "OpenAI model for complex coding and agentic workflows", "family": "gpt-sol", "release_date": "2026-09-22", "last_updated": "2026-09-22", "knowledge_cutoff": "2026-04-20", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-6 Sol as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-6 Sol model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the GPT-6 Sol release of September 22, 2026 as exposed under the gpt-6-sol name: OpenAI's model page lists the default snapshot and the snapshot inventory as gpt-6-sol with no dated snapshot, so applications cannot pin a dated revision through OpenAI, although Azure Foundry documents the version as gpt-6-sol (2026-09-22). OpenAI changed the name in place on September 25, 2026 (an image-encoding fix). GPT-6.1 Sol, which OpenAI's model page calls the newer Sol model, the GPT-5.6 Sol tier, GPT-6 Astra and Luna, the pro reasoning mode, ChatGPT and Codex surfaces, fine-tuned derivatives, and later generations are separate service, application, or model boundaries.", "strengths": [ "OpenAI's changelog pins the September 22, 2026 release, the model page documents modalities, limits, endpoints, and rate limits, and the deprecations table names the model as a recommended replacement", "Azure Foundry documents the model with a dated version and no access request, a second managed path" ], "tradeoffs": [ "OpenAI documents no fixed dated snapshot and changed the name's behavior in place on September 25, 2026, so the record cannot pin behavior the way dated-snapshot records can", "OpenAI's page already points to GPT-6.1 Sol as the newer Sol model, and no developer-published weights are available for self-operation" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.42 }, "evidence": [ { "kind": "web", "label": "Official GPT-6 Sol model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-6-sol", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official API changelog: GPT-6 Sol and GPT-6 Luna release (Sep 22) and image-encoding fix (Sep 25)", "url": "https://developers.openai.com/api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Azure Foundry Models sold by Azure listing gpt-6-sol (2026-09-22)", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Azure OpenAI reasoning-models page: gpt-6-sol availability and version date", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official deprecations page: gpt-6-sol named as a replacement, no retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Using GPT-6 guide placing GPT-6.1 Sol and GPT-6 Luna beside Astra", "url": "https://developers.openai.com/api/docs/guides/latest-model", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/openai/gpt-6-sol.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "cyber", "bio_chem" ], "determination": "precautionary", "scope": "weights", "statement": "Under our Preparedness Framework, we are treating GPT-6 Sol and GPT-6 Luna as High capability in both Cybersecurity and Biological and Chemical domains.", "url": "https://deploymentsafety.openai.com/gpt-6-astra/", "content_sha256": "8f13bca387cc4990574bb106556d3a678b7c17cc6c8df0aa4e83df36a4367a82", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-6-1-sol", "source_id": "openai/gpt-6.1-sol", "name": "GPT-6.1 Sol", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "description": "The GPT-6.1 Sol release of September 29, 2026 for complex coding, computer use, and professional work at a lower cost than GPT-6 Astra, exposed as the gpt-6.1-sol name over OpenAI's managed developer API and Azure Foundry with no documented fixed dated snapshot on OpenAI's page and no downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-6.1 Sol", "description": "Near-Astra performance for complex coding, computer use, and professional work at a lower cost.", "family": "gpt-sol", "release_date": "2026-09-29", "last_updated": "2026-09-29", "knowledge_cutoff": "2026-04-30", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1050000, "input": 922000, "output": 128000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-6.1 Sol as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-6.1 Sol model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the GPT-6.1 Sol release of September 29, 2026 as exposed under the gpt-6.1-sol name: OpenAI's model page lists the default snapshot and the snapshot inventory as gpt-6.1-sol with no dated snapshot, so applications cannot pin a dated revision through OpenAI, although Azure Foundry documents the version as gpt-6.1-sol (2026-09-29). It is a release with its own name, knowledge cutoff, reasoning-effort support, and cached-input price, so GPT-6 Sol is a separate record rather than an earlier snapshot of this one. GPT-6 Astra and Luna, the GPT-5.6 Sol tier, the pro reasoning mode selected with a request parameter, ChatGPT and Codex surfaces, fine-tuned derivatives, and later generations are separate service, application, or model boundaries.", "strengths": [ "OpenAI's changelog pins the September 29, 2026 release, the model page documents modalities, limits, endpoints, and rate limits, and the model catalog lists it as a current choice", "Azure Foundry documents the model with a dated version and no access request, a second managed path" ], "tradeoffs": [ "OpenAI documents no fixed dated snapshot, so the record cannot pin behavior the way dated-snapshot records can, and OpenAI states no retirement or deprecation posture for it", "No developer-published weights are available for self-operation or independent deployment" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Official GPT-6.1 Sol model, modality, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official API changelog: GPT-6.1 Sol release (Sep 29)", "url": "https://developers.openai.com/api/docs/changelog", "verified_at": "2026-10-06" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Azure Foundry Models sold by Azure listing gpt-6.1-sol (2026-09-29)", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Azure OpenAI reasoning-models page: gpt-6.1-sol availability and version date", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official deprecations page checked: no gpt-6.1-sol retirement notice", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Using GPT-6 guide covering GPT-6.1 Sol", "url": "https://developers.openai.com/api/docs/guides/latest-model", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/openai/gpt-6.1-sol.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "Critical", "domains": [ "cyber" ], "determination": "determined", "scope": "weights", "statement": "We determined that GPT-6.1 Sol reaches the Critical threshold for cybersecurity capabilities under the Preparedness Framework.", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/", "content_sha256": "42e6a60e8b4a21ed8242f8c472b8e401bdc6623ee37a92df4ba1bd1fedcf47f5", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-audio", "source_id": null, "name": "GPT-Audio", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-audio", "description": "An audio-in, audio-and-text-out chat model, released as the dated snapshot gpt-audio-2025-08-28 behind the gpt-audio name, served through OpenAI's Chat Completions API and Azure Foundry, now deprecated for removal on January 20, 2027, without downloadable weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-Audio", "description": "OpenAI's first generally available audio model; accepts audio inputs and outputs through the Chat Completions API", "family": null, "release_date": "2025-08-28", "last_updated": null, "knowledge_cutoff": "2023-10-01", "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": true, "structured_output": false, "temperature": null }, "limits": { "context": 128000, "input": null, "output": 16384 }, "reported_open_weights": null, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-Audio as an API service governed by the OpenAI Services Agreement and the OpenAI Policies it incorporates by reference. Those terms grant a non-exclusive right to use the Services, prohibit reverse engineering, and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the GPT-Audio model through OpenAI API Services and incorporated OpenAI Policies", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the gpt-audio release identified by its single documented snapshot, gpt-audio-2025-08-28, which the gpt-audio name resolves to. OpenAI marks the model Deprecated and lists gpt-audio for removal on January 20, 2027, naming gpt-audio-1.5 as the replacement. The gpt-audio-mini sibling (whose name resolves to a different snapshot), gpt-audio-1.5, the realtime and transcription models, the earlier gpt-4o-audio-preview releases, the Realtime API, fine-tuned derivatives, and later generations are separate service or model boundaries.", "strengths": [ "OpenAI documents audio and text as both input and output, the dated snapshot, limits, endpoints, and rate limits on the first-party model page", "The deprecations page gives a notification date, a removal date, and a named replacement, and Azure Foundry lists the same version" ], "tradeoffs": [ "OpenAI has deprecated the model and set removal for January 20, 2027, so the record has a short remaining service life", "No developer-published weights are available for self-operation, and OpenAI's page documents no structured-output support" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.54 }, "evidence": [ { "kind": "web", "label": "Official GPT-Audio model, modality, snapshot, and endpoint documentation", "url": "https://developers.openai.com/api/docs/models/gpt-audio", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official API deprecations page: gpt-audio removal on January 20, 2027", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Azure Foundry Models sold by Azure listing gpt-audio (2025-08-28)", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-audio", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-image-2", "source_id": "openai/gpt-image-2", "name": "GPT-Image-2", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-image-2", "description": "OpenAI's image generation and editing model, accepting text and image inputs and producing image outputs, with a dated snapshot behind the gpt-image-2 ID, offered through OpenAI's developer API and documented Azure Foundry hosting, without downloadable weights.", "model_type": "image_generation_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GPT-Image-2", "description": "Image model for prompt-driven generation, editing, and visual design workflows", "family": "gpt-image", "release_date": "2026-04-21", "last_updated": "2026-04-21", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 0, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides GPT-Image-2 as an API service for business and developer customers under the OpenAI Services Agreement, which applies to use of OpenAI's APIs and assigns ownership of Output to the customer. Those terms govern managed access and do not license model weights or implementation for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "scope": "Business and developer access to the model through OpenAI's APIs, and ownership of Output", "kind": "web_terms", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the GPT-Image-2 release behind the gpt-image-2 ID, whose model page lists the dated snapshot gpt-image-2-2026-04-21. The GPT Image 2.5 Flare and Sunburst releases, the deprecated gpt-image-1, gpt-image-1-mini, and gpt-image-1.5 models, ChatGPT image features, the OpenAI API platform, Azure Foundry hosting of the same model, and fine-tuned or downstream derivatives are separate model, service, or application boundaries.", "strengths": [ "The official model page lists a dated snapshot, gpt-image-2-2026-04-21, so applications can pin a revision", "Microsoft's own Foundry models listing names gpt-image-2 as an image generation model, adding managed serving beyond the first-party API" ], "tradeoffs": [ "No developer-published weights are available for self-operation or independent deployment", "Access, quotas, and lifecycle remain controlled by the managed APIs rather than the model operator" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 4.8 }, "evidence": [ { "kind": "web", "label": "Official GPT-Image-2 model, modality, endpoint, and dated-snapshot documentation", "url": "https://developers.openai.com/api/docs/models/gpt-image-2", "verified_at": "2026-10-06" }, { "kind": "web", "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Microsoft Foundry Models sold by Azure listing naming gpt-image-2", "url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official deprecations page checked: no retirement notice for this model", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/openai/gpt-image-2.toml", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://deploymentsafety.openai.com/chatgpt-images-2-0/", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-oss-120b", "source_id": "openai/gpt-oss-120b", "name": "gpt-oss-120b", "developer": "OpenAI", "url": "https://huggingface.co/openai/gpt-oss-120b", "description": "A developer-published mixture-of-experts language-model release whose downloadable checkpoints are governed by the unmodified Apache-2.0 license with a short accompanying usage statement.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "GPT OSS 120B", "description": "Open GPT reasoning model for self-hosted agents and controllable deployments", "family": "gpt-oss", "release_date": "2025-08-05", "last_updated": "2025-08-05", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-120b" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text and governs the published weights, tokenizer, and chat template. A separate USAGE_POLICY file asks users to comply with applicable law but adds no use-based restrictions beyond the license, so the release is open source.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "openai/gpt-oss-120b model weights, tokenizer, chat template, and original and Metal checkpoint artifacts", "kind": "web_terms", "url": "https://huggingface.co/openai/gpt-oss-120b/blob/b5c939de8f754692c1647ca79fbf85e8c1e70f8a/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the openai/gpt-oss-120b weight release at Hugging Face commit b5c939de8f754692c1647ca79fbf85e8c1e70f8a, last modified 2025-08-26. The gpt-oss-20b sibling, downstream quantizations and fine-tunes, OpenAI's managed API services, and independent hosting providers remain separate boundaries.", "strengths": [ "OpenAI publishes Transformers-layout Safetensors, original checkpoint, and Metal artifacts in one ungated repository with a repository license file", "The model card documents Transformers, vLLM, PyTorch/Triton, Ollama, LM Studio, and SGLang serving paths for self-operation" ], "tradeoffs": [ "The published MXFP4 checkpoint is sized for a single 80 GB accelerator, so self-operation still needs data-centre-class hardware", "Neither the model card nor OpenAI's docs page carries dated release notes or a deprecation posture, so version identity rests on repository commits" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 8.08 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoints, and runtime guidance", "url": "https://huggingface.co/openai/gpt-oss-120b", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official OpenAI docs page for gpt-oss-120b", "url": "https://developers.openai.com/api/docs/models/gpt-oss-120b", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/openai/gpt-oss-120b.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High capability", "domains": [ "bio_chem", "cyber", "autonomy" ], "determination": "below_threshold", "scope": "weights", "statement": "We ran scalable capability evaluations on gpt-oss-120b, and confirmed that the default model does not reach our indicative thresholds for High capability in any of the three Tracked Categories of our Preparedness Framework (Biological and Chemical capability, Cyber capability, and AI Self-Improvement).", "url": "https://deploymentsafety.openai.com/gpt-oss/", "content_sha256": "229bcad586c82893d0834b40ec74ef91def2506efbc30045d70fd41a200da1fc", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-gpt-realtime-2-1", "source_id": "openai/gpt-realtime-2.1", "name": "GPT-Realtime-2.1", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-2.1", "description": "An API-served realtime speech model with audio and text output, available through OpenAI's Realtime API under the OpenAI Services Agreement, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT-Realtime-2.1", "description": "Realtime speech-to-speech model with configurable reasoning, tool use, and robust voice-agent behavior", "family": "gpt", "release_date": "2026-07-06", "last_updated": "2026-07-06", "knowledge_cutoff": "2024-09-30", "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": false }, "limits": { "context": 128000, "input": 96000, "output": 32000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-18", "scope": "Commercial access to GPT-Realtime-2.1 through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the GPT-Realtime-2.1 release of July 2026, which updates GPT-Realtime-2 and is named as the recommended replacement for the retiring gpt-realtime and gpt-4o-realtime lines. The Realtime API platform, the gpt-realtime-2.1-mini sibling, the transcription models, and later OpenAI releases remain separate service or product boundaries.", "strengths": [ "OpenAI documents the model page, launch date, and its placement as the current replacement in the deprecations table", "The modalities, endpoints, and rate limits are documented first-party" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The snapshot identity is alias-only with nothing pinnable, and the release is served only through the Realtime API" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.26 }, "evidence": [ { "kind": "web", "label": "Official GPT-Realtime-2.1 model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-2.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Realtime model release announcement", "url": "https://community.openai.com/t/new-realtime-models-on-the-api-gpt-realtime-2-1-and-gpt-realtime-2-1-mini/1385896", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-realtime-2.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-2.1", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-gpt-realtime-whisper", "source_id": "openai/gpt-realtime-whisper", "name": "GPT-Realtime-Whisper", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-whisper", "description": "An API-served streaming speech-to-text model with text output, available through OpenAI's realtime transcription endpoint under the OpenAI Services Agreement, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GPT Realtime Whisper", "description": "Streaming speech-to-text model for low-latency transcript deltas from live audio", "family": "whisper", "release_date": "2026-05-07", "last_updated": "2026-05-07", "knowledge_cutoff": null, "modalities": { "input": [ "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 0, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-18", "scope": "Commercial access to GPT-Realtime-Whisper through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the GPT-Realtime-Whisper release of May 2026, a streaming speech-to-text model served through the v1/realtime/transcription_sessions endpoint. The deprecated whisper-1 and GPT-4o Transcribe family, the gpt-transcribe and gpt-live-transcribe successors, the Realtime API platform, and later OpenAI releases remain separate service or product boundaries.", "strengths": [ "OpenAI documents the model page, launch date, and the transcription endpoint", "The release is explicitly outside the deprecated transcription family named in the deprecations page" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The snapshot identity is alias-only with nothing pinnable, and the release is served only through one dedicated endpoint" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 4.12 }, "evidence": [ { "kind": "web", "label": "Official GPT-Realtime-Whisper model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-whisper", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official voice intelligence announcement", "url": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/gpt-realtime-whisper.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-whisper", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-o1", "source_id": "openai/o1", "name": "o1", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/o1", "description": "The first o-series reasoning API line, served under the OpenAI Services Agreement without downloadable model weights; OpenAI names GPT-5.6 Sol as its recommended replacement.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "o1", "description": "O-series reasoning model for hard analysis, math, coding, and planning", "family": "o", "release_date": "2024-12-05", "last_updated": "2024-12-05", "knowledge_cutoff": "2023-09", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 200000, "input": null, "output": 100000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the o1 release identified by the dated snapshot o1-2024-12-17, announced December 2024. OpenAI announced its deprecation and documents the shutdown date of October 23, 2026 with GPT-5.6 Sol as the recommended replacement. The o1-pro sibling, the later o3 and o4-mini lines, the GPT-5 family, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "Reasoning behavior, image and PDF input, and structured outputs are documented" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access ends at the documented shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official o1 model documentation", "url": "https://developers.openai.com/api/docs/models/o1", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/o1.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "medium", "domains": [ "bio_chem" ], "determination": "determined", "scope": "weights", "statement": "Our evaluations found that o1 can help experts with the operational planning of reproducing a known biological threat, which meets our medium risk threshold.", "url": "https://cdn.openai.com/o1-system-card-20241205.pdf", "content_sha256": "3ba7bdbe69e022f6be0527edbfe18ac1becdea48d5be9891ab764ff336c8a033", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-o1-pro", "source_id": "openai/o1-pro", "name": "o1-pro", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/o1-pro", "description": "A higher-compute version of the o1 reasoning line, served only through the Responses API under the OpenAI Services Agreement without downloadable model weights; OpenAI recommends the GPT-5.6 Sol line at reasoning.mode pro.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "o1-pro", "description": "O-series reasoning model for hard analysis, math, coding, and planning", "family": "o-pro", "release_date": "2025-03-19", "last_updated": "2025-03-19", "knowledge_cutoff": "2023-09", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 200000, "input": null, "output": 100000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the o1-pro release identified by the dated snapshot o1-pro-2025-03-19, announced March 2025. OpenAI announced its deprecation and documents the shutdown date of October 23, 2026 with the GPT-5.6 Sol line at reasoning.mode pro as the recommended replacement. The o1 line it derives from, the later o3-pro, the GPT-5 family, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "The separate pricing and the Responses-API-only serving path are documented" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Served only through the Responses API with no Chat Completions path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 9, "documentation_provenance": 8, "overall": 4.5 }, "evidence": [ { "kind": "web", "label": "Official o1-pro model documentation", "url": "https://developers.openai.com/api/docs/models/o1-pro", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/o1-pro.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/o1-pro", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-o3", "source_id": "openai/o3", "name": "o3", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/o3", "description": "A deliberate o-series reasoning API line, served under the OpenAI Services Agreement without downloadable model weights; OpenAI names GPT-5.6 Sol as its recommended replacement.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "o3", "description": "Deliberate o-series reasoner for hard math, coding, and multi-step analysis", "family": "o", "release_date": "2025-04-16", "last_updated": "2025-04-16", "knowledge_cutoff": "2024-05", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 200000, "input": null, "output": 100000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the o3 release identified by the dated snapshot o3-2025-04-16, announced April 2025. OpenAI announced its deprecation and documents the shutdown date of December 11, 2026 with GPT-5.6 Sol as the recommended replacement. The o3-mini and o3-pro siblings, the later GPT-5 family, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "Reasoning behavior, image and PDF input, and structured outputs are documented" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access ends at the documented shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official o3 model documentation", "url": "https://developers.openai.com/api/docs/models/o3", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/o3.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "bio_chem", "cyber", "autonomy" ], "determination": "below_threshold", "scope": "weights", "statement": "OpenAI’s Safety Advisory Group (SAG) reviewed the results of our Preparedness evaluations and determined that OpenAI o3 and o4-mini do not reach the High threshold in any of our three Tracked Categories: Biological and Chemical Capability, Cybersecurity, and AI Self-improvement.", "url": "https://deploymentsafety.openai.com/o3/", "content_sha256": "72f5606c7115141dee7736f09a853fa5fb4cda2b5315ad9ca1c43a5ca4310bc6", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-o3-mini", "source_id": "openai/o3-mini", "name": "o3-mini", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/o3-mini", "description": "A smaller o-series reasoning API line, served under the OpenAI Services Agreement without downloadable model weights; OpenAI names GPT-5.6 Sol as its recommended replacement.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "o3-mini", "description": "Smaller o-series reasoner for economical coding, math, and planning tasks", "family": "o-mini", "release_date": "2024-12-20", "last_updated": "2025-01-29", "knowledge_cutoff": "2024-05", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 200000, "input": null, "output": 100000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the o3-mini release identified by the dated snapshot o3-mini-2025-01-31, announced December 2024. OpenAI announced its deprecation and documents the shutdown date of October 23, 2026 with GPT-5.6 Sol as the recommended replacement. The o3 and o3-pro siblings, the later GPT-5 family, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "Reasoning behavior, structured outputs, and tool use are documented" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access ends at the documented shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official o3-mini model documentation", "url": "https://developers.openai.com/api/docs/models/o3-mini", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/o3-mini.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "Medium", "domains": [ "bio_chem", "autonomy" ], "determination": "determined", "scope": "weights", "statement": "Under the Preparedness Framework, OpenAI’s Safety Advisory Group (SAG) recommended classifying the OpenAI o3-mini (Pre-Mitigation) model as Medium risk overall. It scores Medium risk for Persuasion, CBRN (chemical, biological, radiological, nuclear), and Model Autonomy, and Low risk for Cybersecurity.", "url": "https://cdn.openai.com/o3-mini-system-card-feb10.pdf", "content_sha256": "420c260d935429e38a31044b287378802218b6edcc5de72b66cca9346b0197fa", "verified_at": "2026-10-09", "research_confidence": "high" } ] }, { "id": "model-openai-o3-pro", "source_id": "openai/o3-pro", "name": "o3-pro", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/o3-pro", "description": "A higher-compute version of the o3 reasoning line, served only through the Responses API under the OpenAI Services Agreement without downloadable model weights; OpenAI recommends the GPT-5.6 Sol line at reasoning.mode pro.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "o3-pro", "description": "High-effort o3 tier for difficult technical reasoning and careful answers", "family": "o-pro", "release_date": "2025-06-10", "last_updated": "2025-06-10", "knowledge_cutoff": "2024-05", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 200000, "input": null, "output": 100000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the o3-pro release identified by the dated snapshot o3-pro-2025-06-10, announced June 2025. OpenAI announced its deprecation and documents the shutdown date of December 11, 2026 with the GPT-5.6 Sol line at reasoning.mode pro as the recommended replacement. The o3 and o3-mini lines it derives from, the later GPT-5 family, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "The separate pricing and the Responses-API-only serving path are documented" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Served only through the Responses API with no Chat Completions path" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 9, "documentation_provenance": 8, "overall": 4.5 }, "evidence": [ { "kind": "web", "label": "Official o3-pro model documentation", "url": "https://developers.openai.com/api/docs/models/o3-pro", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/o3-pro.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://developers.openai.com/api/docs/models/o3-pro", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openai-o4-mini", "source_id": "openai/o4-mini", "name": "o4-mini", "developer": "OpenAI", "url": "https://developers.openai.com/api/docs/models/o4-mini", "description": "A fast o-series reasoning API line, served under the OpenAI Services Agreement without downloadable model weights; OpenAI names GPT-5.6 Terra as its recommended replacement.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "o4-mini", "description": "Fast o-series model for compact reasoning, coding, and tool use", "family": "o-mini", "release_date": "2025-04-16", "last_updated": "2025-04-16", "knowledge_cutoff": "2024-05", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 200000, "input": null, "output": 100000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenAI-Services-Agreement" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "OpenAI provides the release as a managed service under the OpenAI Services Agreement, which permits customer applications but licenses no model weights for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-OpenAI-Services-Agreement", "kind": "web_terms", "url": "https://openai.com/policies/service-terms/", "verified_at": "2026-09-22", "scope": "Commercial access through the OpenAI API and related OpenAI services" } ], "access_boundary": "This record represents the o4-mini release identified by the dated snapshot o4-mini-2025-04-16, announced April 2025. OpenAI announced its deprecation and documents the shutdown date of October 23, 2026 with GPT-5.6 Terra as the recommended replacement. The o1 and o3 lines, the later GPT-5 family, the OpenAI platform, and ChatGPT remain separate product or service boundaries.", "strengths": [ "OpenAI documents the pinned snapshot, deprecation date, and recommended replacement first-party", "Reasoning behavior, image input, tool use, and structured outputs are documented" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Access ends at the documented shutdown date" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 9, "documentation_provenance": 9, "overall": 4.78 }, "evidence": [ { "kind": "web", "label": "Official o4-mini model documentation", "url": "https://developers.openai.com/api/docs/models/o4-mini", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official API deprecations and model lifecycle page", "url": "https://developers.openai.com/api/docs/deprecations", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openai/o4-mini.toml", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22", "flags": [ { "kind": "maker_risk_safeguards", "status": "statement_found", "tier_term": "High", "domains": [ "bio_chem", "cyber", "autonomy" ], "determination": "below_threshold", "scope": "weights", "statement": "OpenAI’s Safety Advisory Group (SAG) reviewed the results of our Preparedness evaluations and determined that OpenAI o3 and o4-mini do not reach the High threshold in any of our three Tracked Categories: Biological and Chemical Capability, Cybersecurity, and AI Self-improvement.", "url": "https://deploymentsafety.openai.com/o3/", "content_sha256": "72f5606c7115141dee7736f09a853fa5fb4cda2b5315ad9ca1c43a5ca4310bc6", "verified_at": "2026-10-09", "research_confidence": "medium" } ] }, { "id": "model-openbmb-minicpm5-1b", "source_id": "openbmb/minicpm5-1b", "name": "MiniCPM5-1B", "developer": "OpenBMB", "url": "https://huggingface.co/openbmb/MiniCPM5-1B", "description": "OpenBMB's MiniCPM5-1B, the first checkpoint of the MiniCPM5 series, a dense 1-billion-parameter on-device language model with 131K context and hybrid think/no-think modes, published under Apache-2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "MiniCPM5-1B", "description": "Dense 1B-class open-source model for on-device and resource-constrained use, with native long-context support, Think / No Think chat modes, and tool calling", "family": null, "release_date": "2026-05-19", "last_updated": "2026-05-19", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": "apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/openbmb/MiniCPM5-1B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The Hugging Face repository ships no separate LICENSE file, but the OpenBMB/MiniCPM GitHub repository, which is the first-party home of the family, carries the full Apache-2.0 license text, and the model card declares Apache-2.0.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "MiniCPM5-1B weights and checkpoints under the OpenBMB Apache-2.0 family license", "kind": "web_terms", "url": "https://github.com/OpenBMB/MiniCPM/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the MiniCPM5-1B weight release, the first checkpoint of the MiniCPM5 series. The MiniCPM3 and MiniCPM4 generations, later MiniCPM5 checkpoints, quantized and GGUF variants, and hosted services are separate boundaries.", "strengths": [ "Genuine new family checkpoint, light and on-device oriented with a hybrid reasoning mode", "Apache-2.0 grant backed by the first-party OpenBMB/MiniCPM repository license text" ], "tradeoffs": [ "The Hugging Face repository ships no LICENSE file; the grant is evidenced from the GitHub repository", "The 1B scale ceiling limits harder reasoning and coding tasks" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 7.02 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/openbmb/minicpm5-1b.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/openbmb/MiniCPM5-1B", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-openbmb-minicpm5-2b", "source_id": "openbmb/minicpm5-2b", "name": "MiniCPM5-2B", "developer": "OpenBMB", "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "description": "OpenBMB's MiniCPM5-2B, the second checkpoint of the MiniCPM5 series, a dense 2.5-billion-parameter on-device language model with 131K context and tool calling, published under Apache-2.0.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "MiniCPM5-2B", "description": "Dense 2B-class open-source model for on-device and resource-constrained use, with native long-context support, tool calling, and agentic tasks", "family": null, "release_date": "2026-09-06", "last_updated": "2026-09-12", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": "apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/openbmb/MiniCPM5-2B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The Hugging Face repository ships no separate LICENSE file, but the OpenBMB/MiniCPM GitHub repository carries the full Apache-2.0 text, matching the evidence used for the reviewed MiniCPM5-1B record.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "MiniCPM5-2B weights and checkpoints under the OpenBMB Apache-2.0 family license", "kind": "web_terms", "url": "https://github.com/OpenBMB/MiniCPM/blob/main/LICENSE", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the MiniCPM5-2B weight release, the sibling checkpoint of the reviewed MiniCPM5-1B. The MiniCPM5-1B and MiniCPM3/4 generations, quantized and GGUF variants, and hosted services are separate boundaries.", "strengths": [ "Genuine sibling checkpoint of the just-reviewed MiniCPM5-1B with strong coding, math, and agent strengths at 2B scale", "Apache-2.0 grant backed by the first-party OpenBMB/MiniCPM repository license text" ], "tradeoffs": [ "The Hugging Face repository ships no LICENSE file; the grant is evidenced from the GitHub repository", "The 2B scale ceiling limits the hardest reasoning and coding tasks" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 3, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 7.02 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/openbmb/minicpm5-2b.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-perplexity-sonar", "source_id": "perplexity/sonar", "name": "Sonar", "developer": "Perplexity", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar", "description": "Perplexity's lightweight search release for quick grounded answers with real-time web search, with 128K context and no public weights, served as perplexity/sonar on the Agent API after support for the Sonar API ended on September 27, 2026.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Sonar", "description": "Fast web-grounded Sonar for current answers, citations, and lightweight retrieval", "family": "sonar", "release_date": "2024-01-01", "last_updated": "2025-09-01", "knowledge_cutoff": "2025-09-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Perplexity provides these models as API services governed by the Perplexity API Terms of Service (January 2026), which grant a non-exclusive, non-sublicensable, non-transferable right to submit inputs and display outputs while Perplexity retains all intellectual property in the models. No weights or model license exist; the API trains on no customer content.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the sonar model with no published model license", "kind": "web_terms", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the sonar search line, served through the Perplexity Sonar API until Perplexity ended Sonar Chat Completions support on September 27, 2026, and through the Agent API as perplexity/sonar, whose model list links the same Sonar model page. The Pro, Reasoning Pro, and Deep Research siblings, retired Llama-based sonar names, the Agent API presets, and the Perplexity platform itself are separate boundaries.", "strengths": [ "Dedicated first-party page pins the model with 128K context, citations, and search grounding", "Still served on the Agent API as perplexity/sonar, with first-party SDKs and OpenAI-compatible access" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this model", "The evergreen name carries no pinnable snapshot, and its Sonar API route lost support on September 27, 2026, leaving the Agent API as the documented path" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 2, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 4, "documentation_provenance": 9, "overall": 5.16 }, "evidence": [ { "kind": "web", "label": "Official Sonar model documentation with samples", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Sonar-to-Agent-API migration guide recording that Sonar Chat Completions support ended on September 27, 2026", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Agent API model list naming perplexity/sonar and linking the Sonar model page", "url": "https://docs.perplexity.ai/docs/agent-api/models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Perplexity API Terms of Service (January 2026)", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/perplexity/sonar.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-28" }, { "id": "model-perplexity-sonar-deep-research", "source_id": "perplexity/sonar-deep-research", "name": "Sonar Deep Research", "developer": "Perplexity", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar-deep-research", "description": "Perplexity's exhaustive research release generating cited reports over hundreds of sources, served as the sonar-deep-research Sonar API model with 128K context and no public weights until support for that API ended on September 27, 2026.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Sonar Deep Research", "description": "Sonar search model for autonomous research and citation-backed long-form reports", "family": "sonar", "release_date": "2025-02-01", "last_updated": "2025-09-01", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 128000, "input": null, "output": 32768 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Perplexity provides these models as API services governed by the Perplexity API Terms of Service (January 2026), which grant a non-exclusive, non-sublicensable, non-transferable right to submit inputs and display outputs while Perplexity retains all intellectual property in the models. No weights or model license exist; the API trains on no customer content.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the sonar-deep-research model with no published model license", "kind": "web_terms", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the sonar-deep-research line served through the Perplexity Sonar API with asynchronous execution and reasoning effort. Perplexity ended Sonar Chat Completions support on September 27, 2026: synchronous and streaming requests are being reformulated as Agent API requests, rolling out by model, asynchronous requests are no longer supported, and the Agent API lists no sonar-deep-research model, only a migration mapping to its high preset. Other Sonar siblings, the Agent API presets, and the platform itself are separate boundaries.", "strengths": [ "Dedicated first-party page pins the model with citation, search-query, and reasoning billing dimensions", "Documented reasoning effort trades research depth against speed and cost" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this model", "The evergreen name carries no pinnable snapshot, and its API support, including the asynchronous mode its long runs used, ended on September 27, 2026 with no Agent API model to succeed it, only the high preset" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 2, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 5, "documentation_provenance": 9, "overall": 5.14 }, "evidence": [ { "kind": "web", "label": "Official Sonar Deep Research model documentation", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar-deep-research", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Sonar-to-Agent-API migration guide recording that Sonar Chat Completions support ended on September 27, 2026", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Agent API model list, which names perplexity/sonar and no sonar-deep-research model", "url": "https://docs.perplexity.ai/docs/agent-api/models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Perplexity API Terms of Service (January 2026)", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/perplexity/sonar-deep-research.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-28" }, { "id": "model-perplexity-sonar-pro", "source_id": "perplexity/sonar-pro", "name": "Sonar Pro", "developer": "Perplexity", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar-pro", "description": "Perplexity's advanced search release for complex multi-step questions with deeper retrieval, served as the sonar-pro Sonar API model with 200K context and no public weights until support for that API ended on September 27, 2026.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Sonar Pro", "description": "Deeper Sonar search model with broader retrieval and stronger synthesis", "family": "sonar-pro", "release_date": "2024-01-01", "last_updated": "2025-09-01", "knowledge_cutoff": "2025-09-01", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Perplexity provides these models as API services governed by the Perplexity API Terms of Service (January 2026), which grant a non-exclusive, non-sublicensable, non-transferable right to submit inputs and display outputs while Perplexity retains all intellectual property in the models. No weights or model license exist; the API trains on no customer content.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the sonar-pro model with no published model license", "kind": "web_terms", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the sonar-pro search line served through the Perplexity Sonar API. Perplexity ended Sonar Chat Completions support on September 27, 2026: synchronous and streaming requests are being reformulated as Agent API requests, rolling out by model, and the Agent API lists no sonar-pro model, only a migration mapping to its fast preset. The Pro Search mode is a parameter of this model, not a separate release. Other Sonar siblings, the Agent API presets, and the platform itself are separate boundaries.", "strengths": [ "Dedicated first-party page pins the model with 200K context and doubled search results", "Pro Search mode and follow-up support on the first-party API" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this model", "The evergreen name carries no pinnable snapshot, and its API support ended on September 27, 2026 with no Agent API model to succeed it, only the fast preset" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 2, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 4, "documentation_provenance": 9, "overall": 5.16 }, "evidence": [ { "kind": "web", "label": "Official Sonar Pro model documentation", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar-pro", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Sonar-to-Agent-API migration guide recording that Sonar Chat Completions support ended on September 27, 2026", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Agent API model list, which names perplexity/sonar and no sonar-pro model", "url": "https://docs.perplexity.ai/docs/agent-api/models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Perplexity API Terms of Service (January 2026)", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/perplexity/sonar-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-28" }, { "id": "model-perplexity-sonar-reasoning-pro", "source_id": "perplexity/sonar-reasoning-pro", "name": "Sonar Reasoning Pro", "developer": "Perplexity", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar-reasoning-pro", "description": "Perplexity's chain-of-thought reasoning release for complex multi-step tasks, served as the sonar-reasoning-pro Sonar API model with 128K context and no public weights until support for that API ended on September 27, 2026.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Sonar Reasoning Pro", "description": "Web-grounded Sonar for multi-step research questions that need cited reasoning", "family": "sonar-reasoning", "release_date": "2024-01-01", "last_updated": "2025-09-01", "knowledge_cutoff": "2025-09-01", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Perplexity provides these models as API services governed by the Perplexity API Terms of Service (January 2026), which grant a non-exclusive, non-sublicensable, non-transferable right to submit inputs and display outputs while Perplexity retains all intellectual property in the models. No weights or model license exist; the API trains on no customer content.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only distribution of the sonar-reasoning-pro model with no published model license", "kind": "web_terms", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the sonar-reasoning-pro line served through the Perplexity Sonar API, the named successor of the December 2025 deprecated sonar-reasoning model. Perplexity ended Sonar Chat Completions support on September 27, 2026: synchronous and streaming requests are being reformulated as Agent API requests, rolling out by model, and the Agent API lists no sonar-reasoning-pro model, only a migration mapping to its low preset. Other Sonar siblings, the Agent API presets, and the platform itself are separate boundaries.", "strengths": [ "Dedicated first-party page pins the model with chain-of-thought reasoning and structured outputs", "Explicit predecessor-deprecation lineage on first-party docs" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this model", "The evergreen name carries no pinnable snapshot, and its API support ended on September 27, 2026 with no Agent API model to succeed it, only the low preset" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 2, "deployment_portability": 2, "serving_reach": 8, "lifecycle_transparency": 5, "documentation_provenance": 9, "overall": 5.28 }, "evidence": [ { "kind": "web", "label": "Official Sonar Reasoning Pro model documentation", "url": "https://docs.perplexity.ai/docs/sonar/models/sonar-reasoning-pro", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Sonar-to-Agent-API migration guide recording that Sonar Chat Completions support ended on September 27, 2026", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Agent API model list, which names perplexity/sonar and no sonar-reasoning-pro model", "url": "https://docs.perplexity.ai/docs/agent-api/models", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Perplexity API Terms of Service (January 2026)", "url": "https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/perplexity/sonar-reasoning-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-28" }, { "id": "model-poolside-laguna-m-1", "source_id": "poolside/laguna-m.1", "name": "Laguna M.1", "developer": "Poolside", "url": "https://huggingface.co/poolside/Laguna-M.1", "description": "Poolside's 225-billion-parameter flagship release with 23B active parameters, published as Apache-2.0 open weights with managed API serving.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Laguna M.1", "description": "Poolside's open-weight model for agentic coding and long-horizon work", "family": "laguna", "release_date": "2026-04-28", "last_updated": "2026-06-13", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.md file is the verbatim Apache License 2.0 text with no use restrictions attached, so the release is open source. The model card asks that use stay consistent with Poolside's Acceptable Use Policy as companion guidance.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Laguna-M.1 model weights and materials in the official repository", "kind": "web_terms", "url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/LICENSE.md", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the April 2026 Laguna M.1 release at Hugging Face commit 2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6 (last modified 2026-07-14), covering base, BF16, FP8, and NVFP4 variants, together with the poolside/laguna-m.1 API model. Later Laguna generations, the Poolside Chat product, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 225B flagship checkpoint with base, FP8, and NVFP4 variants under Apache-2.0", "First-party inference API with platform key issuance" ], "tradeoffs": [ "225B total parameters need multi-GPU serving for self-operation", "Version identity rests on repository commits with base, post-trained, and quant variants to spell out" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.86 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/poolside/Laguna-M.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Laguna deeper-dive announcement", "url": "https://poolside.ai/blog/laguna-a-deeper-dive", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/poolside/laguna-m.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-poolside-laguna-s-2-1", "source_id": "poolside/laguna-s-2.1", "name": "Laguna S 2.1", "developer": "Poolside", "url": "https://huggingface.co/poolside/Laguna-S-2.1", "description": "Poolside's 118-billion-parameter release with 8B active parameters and 1M context, published as OpenMDW-1.1 open weights with managed API and third-party hosting.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Laguna S 2.1", "description": "Agentic coding model from Poolside in the XS size class for local deployment", "family": "laguna", "release_date": "2026-07-21", "last_updated": "2026-07-21", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.md file is the OpenMDW License Agreement 1.1, which grants dealing in the model materials without restriction and imposes no obligations on outputs, so the release is open source. The model card references Poolside's Acceptable Use Policy as companion guidance.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "Laguna-S-2.1 model weights and materials in the official repository", "kind": "web_terms", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/LICENSE.md", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the July 2026 Laguna S 2.1 release at Hugging Face commit 0f573140834b11cfac0c2af97a101a7a69a13e22 (last modified 2026-08-19), together with the poolside/laguna-s-2.1 API model and its OpenRouter hosting. Other Laguna generations, the Poolside Chat product, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 118B checkpoint with BF16, FP8, NVFP4, INT4, GGUF, and DFlash variants under OpenMDW-1.1", "First-party API plus exact-model OpenRouter hosting with 1M context" ], "tradeoffs": [ "Version identity rests on repository commits with no snapshot versioning", "The OpenMDW identifier is new to the catalog and needs its first review cycle" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.1 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/poolside/Laguna-S-2.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "OpenRouter model page for the exact poolside/laguna-s-2.1 model", "url": "https://openrouter.ai/poolside/laguna-s-2.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/poolside/laguna-s-2.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-poolside-laguna-xs-2-1", "source_id": "poolside/laguna-xs-2.1", "name": "Laguna XS 2.1", "developer": "Poolside", "url": "https://huggingface.co/poolside/Laguna-XS-2.1", "description": "Poolside's 33-billion-parameter efficient release with 3B active parameters for local agentic coding, published as OpenMDW-1.1 open weights with managed API and third-party hosting.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Laguna XS 2.1", "description": "Agentic coding model from Poolside in the XS size class for local deployment", "family": "laguna", "release_date": "2026-07-02", "last_updated": "2026-07-02", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-OpenMDW-1.1" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.md file is the OpenMDW License Agreement 1.1, byte-identical to the S 2.1 grant, imposing no restrictions or output obligations, so the release is open source.", "license_evidence": [ { "license_id": "LicenseRef-OpenMDW-1.1", "scope": "Laguna-XS-2.1 model weights and materials in the official repository", "kind": "web_terms", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/LICENSE.md", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the June 2026 Laguna XS 2.1 release at Hugging Face commit c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db (last modified 2026-08-19), an upgrade of the XS.2 line rather than a snapshot of it, together with its API model and OpenRouter hosting. The XS.2 predecessor, other Laguna generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 33B checkpoint under OpenMDW-1.1, compact enough for high-RAM local machines", "First-party API plus exact-model OpenRouter hosting at 256K context" ], "tradeoffs": [ "The upgrade-over-XS.2 positioning must be spelled out so the similar names are not conflated", "Version identity rests on repository commits with no snapshot versioning" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.18 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/poolside/Laguna-XS-2.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "OpenRouter model page for the exact poolside/laguna-xs-2.1 model", "url": "https://openrouter.ai/poolside/laguna-xs-2.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/poolside/laguna-xs-2.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-poolside-laguna-xs-2", "source_id": "poolside/laguna-xs.2", "name": "Laguna XS.2", "developer": "Poolside", "url": "https://huggingface.co/poolside/Laguna-XS.2", "description": "Poolside's 33-billion-parameter efficient release, published as Apache-2.0 open weights with local runtimes after its API sunset.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Laguna XS.2", "description": "Agentic coding model from Poolside in the XS size class for local deployment", "family": "laguna", "release_date": "2026-04-28", "last_updated": "2026-06-13", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": false, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.md file is the verbatim Apache License 2.0 text with no use restrictions attached, so the release is open source.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Laguna-XS.2 model weights and materials in the official repository", "kind": "web_terms", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/69e3f4046616e40fb55ac54e0e2e6accbe5cadfe/LICENSE.md", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the April 2026 Laguna XS.2 release at Hugging Face commit 69e3f4046616e40fb55ac54e0e2e6accbe5cadfe (last modified 2026-07-14). Its API line sunset one week after the XS 2.1 launch. The XS 2.1 successor, other Laguna generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 33B checkpoint carrying a verbatim Apache-2.0 license file beside the weights", "One-command Ollama distribution for local operation" ], "tradeoffs": [ "The managed API sunset leaves self-operation as the supported path", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.46 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/poolside/Laguna-XS.2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/poolside/laguna-xs.2.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-prism-ml-ternary-bonsai-2-27b", "source_id": null, "name": "Ternary Bonsai 2 27B", "developer": "Prism ML", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf", "description": "Prism ML's Bonsai 2 27B, a ternary-weight (1.72 bits per weight) build of Alibaba's Qwen3.8-27B that fits in about 5.95 GB, keeps the 262,144-token context and image input, and ships as Apache-2.0 GGUF files that need Prism ML's llama.cpp fork.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Ternary Bonsai 2 27B", "description": "Ternary-weight build of Qwen3.8-27B for on-device reasoning, tool calling, and image understanding with a 262K-token context window.", "family": "bonsai", "release_date": "2026-09-17", "last_updated": "2026-09-25", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [ { "label": "Announcement", "url": "https://prismml.com/news/bonsai-2-27b" }, { "label": "Whitepaper", "url": "https://github.com/PrismML-Eng/Bonsai-demo/blob/main/bonsai-2-27b-whitepaper.pdf" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file is the unmodified Apache License 2.0 text, the model card front matter declares apache-2.0, and NOTICE.txt states the software is copyright 2026-present Prism ML, Inc. under Apache 2.0, discloses that it is built from Qwen3.8-27B (Copyright 2026 Alibaba Cloud, Apache 2.0), and asks for attribution only as a courtesy ('would appreciate'), not as a condition. Both grants are permissive with no use restrictions, so the release is open source with both provenances recorded. The Prism ML llama.cpp and MLX forks that the weights need are separate repositories and are not covered by this review.", "license_evidence": [ { "kind": "web_terms", "license_id": "Apache-2.0", "scope": "Ternary-Bonsai-2-27B GGUF weights and mmproj files under the repository Apache License 2.0 file", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/blob/b072e1d3b35a0a630cece372c2127528e0994386/LICENSE", "verified_at": "2026-10-06" }, { "kind": "web_terms", "license_id": "Apache-2.0", "scope": "Prism ML copyright notice and Qwen3.8-27B Apache-2.0 upstream provenance in the repository NOTICE file", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/blob/b072e1d3b35a0a630cece372c2127528e0994386/NOTICE.txt", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the prism-ml/Ternary-Bonsai-2-27B-gguf weight release announced 17 September 2026, at Hugging Face commit b072e1d3b35a0a630cece372c2127528e0994386 (last modified 2026-09-25): the PTQ1_0 and PQ2_0 ternary GGUF packings, the F16 reference GGUF, and the optional mmproj vision files. The earlier Ternary Bonsai 27B and Bonsai 27B releases, the Bonsai 8B, 4B, and 1.7B and Bonsai Image families, the MLX 2-bit companion repository, community derivatives and quantizations, the Qwen3.8-27B base model, and every hosted route for the weights, including OpenRouter's listing, are separate boundaries.", "strengths": [ "Developer-published Apache-2.0 weights at about 5.95 GB for a 27B-class multimodal reasoning model, with an F16 reference file and Q8_0 and BF16 vision projectors in the same repository", "A whitepaper, a model card with per-benchmark tables against FP16, UD-Q4_K_XL, and IQ2_XXS baselines, and a dated known-issues page give the release unusually specific first-party documentation", "Keeps the base model's 262,144-token context and hybrid-attention layout, with a stated 47 tokens per second on an Apple M5 Max and about 130 on an RTX 5090" ], "tradeoffs": [ "The ternary PTQ1_0 and PQ2_0 formats load only in Prism ML's llama.cpp fork (CUDA, Metal, CPU); the model card says stock llama.cpp cannot load them, so portability depends on a vendor-maintained runtime", "The developer's pages document local runtimes only: no first-party managed API was found, and the OpenRouter listing is a third-party route outside this record", "Derived from Qwen3.8-27B with the architecture unchanged, so its release identity is a compression of another lab's model; the known-issues page lists runaway-reasoning, reasoning-effort, and AVX-512 crash problems with open or in-review status" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.6 }, "evidence": [ { "kind": "web", "label": "Official Hugging Face repository and model card for the Bonsai 2 27B GGUF release", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Apache License 2.0 file at the reviewed commit", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/blob/b072e1d3b35a0a630cece372c2127528e0994386/LICENSE", "verified_at": "2026-10-06" }, { "kind": "web", "label": "NOTICE file naming Prism ML's copyright and the Qwen3.8-27B Apache-2.0 upstream", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/blob/b072e1d3b35a0a630cece372c2127528e0994386/NOTICE.txt", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Known-issues page: runtime requirements, workarounds, and fix status", "url": "https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/blob/b072e1d3b35a0a630cece372c2127528e0994386/KNOWN_ISSUES.md", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Prism ML announcement of Bonsai 2 27B", "url": "https://prismml.com/news/bonsai-2-27b", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Bonsai 2 27B whitepaper in the official Bonsai-demo repository", "url": "https://github.com/PrismML-Eng/Bonsai-demo/blob/main/bonsai-2-27b-whitepaper.pdf", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Qwen3.8-27B base model card carrying the Apache-2.0 declaration", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-alibaba-qwen2-5-coder-0-5b", "source_id": "alibaba/qwen2.5-coder-0.5b", "name": "Qwen2.5-Coder-0.5B", "developer": "Qwen", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B", "description": "A compact code-focused Qwen release with developer-published Safetensors weights and an Apache-2.0 distribution boundary.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen2.5-Coder-0.5B", "description": "Tiny open Qwen code model for lightweight completion and on-device coding", "family": "qwen", "release_date": "2024-11-12", "last_updated": "2024-11-12", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The official model repository applies Apache-2.0 to the published model files and weights. The classification describes distribution terms, not model quality or training-data provenance.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Official Qwen2.5-Coder-0.5B model repository, weights, tokenizer, and accompanying files", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B/blob/main/LICENSE", "verified_at": "2026-09-04" } ], "access_boundary": "This record represents the provider-independent Qwen2.5-Coder-0.5B model release. Hosted copies, quantized repackagings, inference providers, and applications built on the model are separate boundaries.", "strengths": [ "Small developer-published artifact with a permissive, familiar license", "Official model card documents direct Transformers loading and Safetensors distribution" ], "tradeoffs": [ "The source record establishes model capabilities and limits but does not establish every downstream runtime or quantization", "A compact specialist release should not be treated as representative of the larger Qwen family" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.96 }, "evidence": [ { "kind": "web", "label": "Official model card and weight repository", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B", "verified_at": "2026-09-04" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/ef71a4865a5821d8cd24f171a0100156a02e808e/models/alibaba/qwen2.5-coder-0.5b.toml", "verified_at": "2026-09-04" } ], "metadata_verified_at": "2026-09-04", "verified_at": "2026-09-04" }, { "id": "model-alibaba-qwen3-235b-a22b-instruct-2507", "source_id": "alibaba/qwen3-235b-a22b-instruct-2507", "name": "Qwen3 235B-A22B Instruct 2507", "developer": "Qwen", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507", "description": "A downloadable text-only instruction update of the Qwen3-235B-A22B mixture-of-experts model, published by Qwen as BF16 Safetensors shards with a standard Apache-2.0 license file in the repository.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Qwen3 235B-A22B Instruct 2507", "description": "Updated large open Qwen3 MoE instruct model for multilingual chat, coding, and tool use", "family": "qwen", "release_date": "2025-07-21", "last_updated": "2025-07-21", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": "Apache 2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository carries an unmodified Apache License 2.0 text (copyright Alibaba Cloud) in a LICENSE file, and the model-card metadata links to that file as the governing license for the weights, tokenizer, and configuration. No acceptable-use policy or additional restrictions are incorporated.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Qwen3-235B-A22B-Instruct-2507 repository, Safetensors weights, tokenizer, and configuration files", "kind": "web_terms", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507/blob/ac9c66cc9b46af7306746a9250f23d47083d689e/LICENSE", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the Qwen/Qwen3-235B-A22B-Instruct-2507 release, the July 2025 non-thinking instruct update of Qwen3-235B-A22B. The original Qwen3-235B-A22B, the Thinking-2507 variant, FP8 and other quantized repackagings, smaller Qwen3 sizes, Qwen Chat, and any hosted Qwen API remain separate boundaries.", "strengths": [ "Ungated BF16 Safetensors shards with an index, tokenizer, and a standard Apache-2.0 license file are published directly by the developer", "The model card documents versioned Transformers, vLLM, and SGLang paths plus a separate configuration for extended-context serving" ], "tradeoffs": [ "The model card names no first-party API or independent hosting path for this release, so served access must be arranged independently", "The full-precision artifact is very large and local-runtime support is stated for the Qwen3 family rather than documented per release" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.68 }, "evidence": [ { "kind": "web", "label": "Official model card, weight repository, and deployment guidance", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Qwen documentation with deployment and run-locally guides", "url": "https://qwen.readthedocs.io/en/latest/", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/alibaba/qwen3-235b-a22b-instruct-2507.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-reflection-beam", "source_id": null, "name": "Beam", "developer": "Reflection AI", "url": "https://reflection.ai/blog/introducing-beam", "description": "Reflection AI's first model: a text-only, 501-billion-parameter sparse Mixture-of-Experts model (23 billion active) for coding, reasoning, and agentic work, announced October 5, 2026 in limited early access with Apache 2.0 weights promised later in October.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Beam", "description": "Sparse Mixture-of-Experts text model with 501 billion total and 23 billion active parameters for coding, reasoning, and agentic workloads", "family": "beam", "release_date": "2026-10-05", "last_updated": "2026-10-05", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": null }, "limits": { "context": 1000000, "input": null, "output": null }, "reported_open_weights": null, "reported_license": null, "links": [ { "label": "Announcement", "url": "https://reflection.ai/blog/introducing-beam" } ], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "At review time no weights are published and no weights license text exists: the announcement says the weights will be released under an Apache 2.0 license later this month, and early access is limited to a waitlisted group on terms the announcement does not publish. The release is classified proprietary today. Re-review when the weights and their license text ship.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "Limited early access to Beam through a waitlist; no weights or weights license published as of the review date, and the announced Apache 2.0 license is a stated intention only", "kind": "web_terms", "url": "https://reflection.ai/blog/introducing-beam", "verified_at": "2026-10-07" } ], "access_boundary": "This record represents Beam as announced on 2026-10-05: the early version made available to a select, waitlisted group of users, not a generally available API or published weights. The weights, technical report, model card, and developer artifacts promised for later in October 2026, the distribution partners and open-source integrations promised at launch, the later models in the series, and downstream quantizations remain separate boundaries. The announcement says Beam is undergoing final red-teaming and evaluations, and gives 256K tokens as the maximum context length during reinforcement learning and 1M tokens as the effective context length after midtraining.", "strengths": [ "First-party announcement published on release day with architecture, training-data and compute figures, a benchmark table naming its comparison models and the method of its efficiency comparison", "Reflection commits to releasing weights under Apache 2.0 together with documentation and a full stack for running, evaluating, and fine-tuning the model" ], "tradeoffs": [ "Early access only: no weights, license text, model card, technical report, or third-party hosting exist yet, so artifact availability and portability rest on a promise", "The announcement says final red-teaming and evaluations are still under way, and it publishes no model identifier, API model page, or pricing" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 1, "lifecycle_transparency": 4, "documentation_provenance": 6, "overall": 2.7 }, "evidence": [ { "kind": "web", "label": "Official Beam announcement: architecture, early-access status, and the Apache 2.0 weights commitment", "url": "https://reflection.ai/blog/introducing-beam", "verified_at": "2026-10-07" }, { "kind": "web", "label": "Reflection Open Safety Framework page: framework text that names no model", "url": "https://reflection.ai/safety", "verified_at": "2026-10-07" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://reflection.ai/blog/introducing-beam", "verified_at": "2026-10-07", "research_confidence": "medium" } ], "metadata_verified_at": "2026-10-07", "verified_at": "2026-10-07" }, { "id": "model-rekaai-reka-flash-3", "source_id": "rekaai/reka-flash-3", "name": "Reka Flash 3", "developer": "Reka", "url": "https://docs.reka.ai/chat/models", "description": "Reka's March 2025 Flash text release with 64K context, tool use, structured outputs, and reasoning, served on Reka-operated infrastructure through the OpenAI-compatible Reka API under the Reka Terms of Use. No weights are published for this checkpoint. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Reka Flash 3", "description": "Efficient model for low-latency assistance, extraction, and routine automation", "family": "reka", "release_date": "2025-03-12", "last_updated": "2025-03-12", "knowledge_cutoff": "2025-01-31", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": false, "structured_output": true, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 58982 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Reka-Terms-of-Use" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "The Reka Terms of Use, last updated September 2026, govern API access to Reka's models: use requires an account with billing details, commercial use runs under the Business Terms, Reka may use customer content to train its models, and disputes go to binding arbitration with class-action waiver. It grants no licence to model weights for redistribution or self-operation.", "license_evidence": [ { "license_id": "LicenseRef-Reka-Terms-of-Use", "scope": "API access to Reka Flash 3 through Reka-operated inference infrastructure", "kind": "web_terms", "url": "https://reka.ai/legal/terms-of-use", "verified_at": "2026-10-14" } ], "access_boundary": "This record represents the reka-flash-3 API release. The Reka Edge and Core lines, the multimodal Vision services, third-party open models served through the same API, and weight checkpoints some other line publishes are separate boundaries.", "strengths": [ "First-party API docs name reka-flash-3 with its context, modalities, and feature flags, served on infrastructure Reka operates", "OpenAI-compatible endpoint with structured outputs and tool use, documented with copy-paste client examples" ], "tradeoffs": [ "No weights are published, so self-operation is unavailable and the API list can retire the model without a stated notice period", "The terms let Reka train on customer content and send disputes to individual arbitration, which enterprise buyers must weigh" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 5, "documentation_provenance": 8, "overall": 4.16 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/59aa52dbc09727fa6ead433b0285970ebc05ba82/models/rekaai/reka-flash-3.toml", "verified_at": "2026-10-14" }, { "kind": "web", "label": "Reka API model list: reka-flash-3 with 64K text context on Reka-operated infrastructure", "url": "https://docs.reka.ai/chat/models", "verified_at": "2026-10-14" }, { "kind": "web", "label": "Reka Terms of Use governing API access to Reka models", "url": "https://reka.ai/legal/terms-of-use", "verified_at": "2026-10-14" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://docs.reka.ai/chat/models", "verified_at": "2026-10-14", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-14", "verified_at": "2026-10-14" }, { "id": "model-sakana-sakana-namazu-v1-0", "source_id": null, "name": "Sakana Namazu v1.0", "developer": "Sakana AI", "url": "https://console.sakana.ai/models?model=sakana-namazu", "description": "Sakana AI's Japanese-specialized language model, served as `sakana-namazu-v1.0` through its own OpenAI-compatible API with a 256K context window, image and document input, and built-in web search and code execution; it is post-trained by Sakana from the open Kimi K2.6 weights and no weights are published.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Sakana Namazu v1.0", "description": "A Japanese-specialized LLM that combines a deep understanding of Japanese culture and business customs with high-performance reasoning, fine-tuned with Sakana AI's in-house data for Japanese language and business workflows", "family": "sakana-namazu", "release_date": "2026-08-03", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 256000, "input": null, "output": null }, "reported_open_weights": false, "reported_license": null, "links": [ { "label": "Official product page", "url": "https://sakana.ai/namazu/" } ], "weights": [] }, "licenses": [ "LicenseRef-Sakana-API-ToS" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No weights are published and no model licence exists; the release is served only as an API under the Sakana AI API Platform Terms of Service, effective 28 September 2026, agreed with Sakana AI Co., Ltd. and covering Sakana Namazu among the services the Company provides through its API. The terms grant no right to reverse engineer the service and prohibit using the service or its contents to generate, train, fine-tune, or otherwise develop another machine learning model, including by distillation or extraction, which is what forecloses recovering the post-training from the endpoint. They govern output separately from weights: as between the parties the user owns its outputs, while the user grants the Company a perpetual, irrevocable, royalty-free, sublicensable licence to its content, and content is used for model training by default with a non-retrospective opt-out in the console. The service is unavailable in the EU, EEA, UK, and Switzerland pending GDPR work, and the terms set Japanese law with exclusive Tokyo District Court jurisdiction for users in Japan and California law with AAA arbitration seated in Los Angeles for everyone else. A separately incorporated Usage Policy, effective 21 July 2026, is incorporated by reference and restricts surveillance, CBRNE, cyberattack, election-interference, and unapproved security-evaluation uses.", "license_evidence": [ { "kind": "web_terms", "license_id": "LicenseRef-Sakana-API-ToS", "scope": "API access to Sakana Namazu v1.0, including the prohibition on reverse engineering and on training or distilling another model from the service, the content licence the user grants, the default training use of content and its opt-out, the regional exclusions, and the governing-law and dispute-resolution terms", "url": "https://console.sakana.ai/terms-of-service", "verified_at": "2026-09-28" }, { "kind": "web_terms", "license_id": "LicenseRef-Sakana-API-ToS", "scope": "The Usage Policy incorporated into the Terms of Service, restricting surveillance, biometric and profiling uses, CBRN and weapons, illegal activity, cyberattack, self-harm, non-consensual sexual content, minors, election interference, and experimental security evaluation without the Company's written approval", "url": "https://console.sakana.ai/usage-policy", "verified_at": "2026-09-28" } ], "access_boundary": "This record represents the `sakana-namazu-v1.0` release, the fixed model ID Sakana AI's console documents as the current version of its Japanese-specialized line, launched with the API announcement of 3 August 2026. The `sakana-namazu` alias, which the console states points to the latest version and currently resolves to v1.0, is a moving name rather than a second release. The Fugu, Fugu Ultra, Fugu Max, and Fugu Cyber models are separate multi-agent orchestration releases that Sakana explicitly contrasts with this one, the March 2026 Namazu alpha prototypes built on DeepSeek-V3.1-Terminus, Llama 3.1, and gpt-oss-120B are a superseded lineage, the underlying Kimi K2.6 weights are Moonshot AI's release and not Sakana AI's, and the LEAP-style on-premises customization stack Sakana sells is a separate product boundary.", "strengths": [ "Sakana AI documents a versioned model ID, the alias that moves to it, a 256K context window, image input by URL or base64, document input to 4 MB, and structured output with text, json_object, and json_schema", "The release is a single in-house model rather than an orchestration layer, which Sakana's own console states directly when it contrasts Namazu with Fugu", "Governing terms, a separate usage policy, and a published per-token price page are all first-party and dated, and the API is documented across Chat Completions, Responses, and Anthropic Messages styles with built-in web search and code execution" ], "tradeoffs": [ "No weights are published and the terms prohibit reverse engineering and any training or distillation from the service, so the post-training cannot be inspected or reproduced", "Content is used for model training by default and the opt-out is not retroactive, while the user grants the Company a perpetual, irrevocable, sublicensable licence to its content", "The service is not available in the EU, EEA, UK, or Switzerland while Sakana works toward GDPR compliance, and the model ID is versioned without a published deprecation or retirement policy for the line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 1, "deployment_portability": 1, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 4.38 }, "evidence": [ { "kind": "web", "label": "Official console model page: the sakana-namazu-v1.0 model ID, the moving alias, the 256K context window, image and file input, structured output, and the contrast with Fugu", "url": "https://console.sakana.ai/models?model=sakana-namazu", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Namazu product page: a Japanese-specialized LLM built on the open Kimi K2.6 and tuned with Sakana AI's in-house data, with the EU, EEA, UK, and Switzerland exclusion", "url": "https://sakana.ai/namazu/", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Sakana AI API Platform Terms of Service, effective 28 September 2026, naming Sakana AI Co., Ltd. and governing the API including Sakana Namazu", "url": "https://console.sakana.ai/terms-of-service", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official Usage Policy incorporated into the terms, effective 21 July 2026", "url": "https://console.sakana.ai/usage-policy", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official get-started page: the api.sakana.ai endpoint, the OpenAI SDK base URL, and the Chat Completions, Responses, and Anthropic Messages request styles with built-in web search and code execution", "url": "https://console.sakana.ai/get-started", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Official pricing page: Sakana Namazu billed pay-as-you-go outside the subscription plans", "url": "https://console.sakana.ai/pricing", "verified_at": "2026-09-28" } ], "metadata_verified_at": "2026-09-28", "verified_at": "2026-09-28" }, { "id": "model-internlm-internlm3-8b-instruct", "source_id": null, "name": "InternLM3 8B Instruct", "developer": "Shanghai AI Laboratory", "url": "https://huggingface.co/internlm/internlm3-8b-instruct", "description": "The Shanghai AI Laboratory's January 2025 InternLM3 instruction release at 8B parameters with deep-thinking and normal modes, 32K context, and tool support, published under Apache 2.0 with weights on the lab's Hugging Face organization. No developer API serves this checkpoint. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "InternLM3 8B Instruct", "description": "8B instruction model for general-purpose use and advanced reasoning, trained on 4 trillion tokens with deep-thinking and normal response modes", "family": "internlm", "release_date": "2025-01-15", "last_updated": "2025-01-15", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 32768, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/internlm/internlm3-8b-instruct" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The internlm/internlm3-8b-instruct model card declares license apache-2.0 and states that code and model weights are licensed under Apache-2.0, and the repository carries an Apache License, Version 2.0 LICENSE.txt file. No separate model-use terms were found on the pages read.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "InternLM3-8B-Instruct code and weights published in the lab's Hugging Face repository", "kind": "web_terms", "url": "https://huggingface.co/internlm/internlm3-8b-instruct/blob/28c99415adaf61767bd1c619f4f99f308fdfd223/LICENSE.txt", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the InternLM3-8B-Instruct checkpoint of the InternLM line. The base model, the InternLM2 and 2.5 generations, the newer Intern-S and Atria releases, and the LMDeploy serving toolkit are separate boundaries.", "strengths": [ "Apache 2.0 grant stated in the model card and carried in a LICENSE.txt file inside the weight repository", "Deep-thinking and normal response modes with tool support in an 8B checkpoint that runs on a single accelerator" ], "tradeoffs": [ "No developer API or managed host serves this checkpoint, so running it is entirely self-operated", "The card's limitations section warns the model may produce biased or harmful content and disclaims responsibility for its dissemination" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 9, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.34 }, "evidence": [ { "kind": "web", "label": "InternLM InternLM3 weight repository: apache-2.0 licence, 8B instruct checkpoint", "url": "https://huggingface.co/internlm/internlm3-8b-instruct", "verified_at": "2026-10-13" }, { "kind": "web", "label": "InternLM project repository: official InternLM series release with Apache-2.0 licence", "url": "https://github.com/InternLM/InternLM", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/internlm/internlm3-8b-instruct", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-xai-grok-4-20-multi-agent", "source_id": null, "name": "Grok 4.20 Multi-Agent", "developer": "SpaceXAI (xAI)", "url": "https://docs.x.ai/developers/models/grok-4.20-multi-agent", "description": "xAI's Grok 4.20 Multi-Agent, a beta text-and-image-input model with a 1,000,000-token context that runs 4 or 16 collaborating agents on research tasks, served through the Grok API under the dated name grok-4.20-multi-agent-0309.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Grok 4.20 Multi-Agent", "description": "Multiple agents collaborate in parallel to perform deep research tasks", "family": "grok", "release_date": "2026-03-10", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 1000000, "input": null, "output": null }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Grok 4.20 Multi-Agent is served, not distributed, under the xAI enterprise terms and acceptable use policy that already govern the reviewed Grok records. The policy forbids using the service or its output to develop competing machine learning models and forbids distilling model data or outputs. No weights are released.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "scope": "Business access to Grok 4.20 Multi-Agent through the SpaceXAI API and related services", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-xAI-AUP", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into xAI service access", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents Grok 4.20 Multi-Agent under the dated API model name grok-4.20-multi-agent-0309, with the grok-4.20-multi-agent name as its alias, as announced with Grok 4.20 on March 10, 2026. The reviewed Grok 4.20 reasoning and non-reasoning snapshots, later Grok releases, and the multi-agent beta feature as a product surface are separate boundaries.", "strengths": [ "A dedicated model page names the dated model name, its aliases, the context window, regions, and rate limits", "A separate capability guide documents the 4-agent and 16-agent configurations and which output is returned" ], "tradeoffs": [ "The model page and capability guide both carry a beta label, and the page lists several beta and experimental aliases", "Closed weights served only through the xAI API under terms that bar competing-model development and distillation" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 4, "documentation_provenance": 6, "overall": 3.62 }, "evidence": [ { "kind": "web", "label": "Official model page for grok-4.20-multi-agent-0309 with its aliases", "url": "https://docs.x.ai/developers/models/grok-4.20-multi-agent", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Multi-agent capability guide documenting the agent counts and beta status", "url": "https://docs.x.ai/developers/model-capabilities/text/multi-agent", "verified_at": "2026-10-06" }, { "kind": "web", "label": "xAI release notes announcing Grok 4.20 and Grok 4.20 Multi-agent on March 10", "url": "https://docs.x.ai/developers/release-notes", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-xai-grok-4-7", "source_id": "xai/grok-4.7", "name": "Grok 4.7", "developer": "SpaceXAI (xAI)", "url": "https://docs.x.ai/developers/models/grok-4.7", "description": "xAI's Grok 4.7, released September 21, 2026, a text-and-image model with 500K context and configurable reasoning effort, served through the Grok API, Grok Build, and several independent hosts.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Grok 4.7", "description": "xAI's frontier model for long-running agents, coding, knowledge work, and visual projects", "family": "grok", "release_date": "2026-09-21", "last_updated": "2026-09-21", "knowledge_cutoff": "2026-05", "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 500000, "input": null, "output": 500000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Grok 4.7 is served, not distributed, under the xAI enterprise terms and acceptable use policy that already govern the reviewed Grok 4.6 record. No weights are released.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "scope": "Business access to Grok 4.7 through the xAI API and related xAI services", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-22" }, { "license_id": "LicenseRef-xAI-AUP", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into xAI service access", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Grok 4.7 release of September 21, 2026, with the pinned grok-4.7 API identifier. The reviewed Grok 4.6 and Grok 4.1 lines, the Grok Voice Transcribe line, and independent hosts are separate boundaries.", "strengths": [ "Dated first-party announcement and full docs model page with pricing, rate limits, and regions", "Distinct new base model independent of the reviewed Grok 4.6 line, with a documented safeguard stack" ], "tradeoffs": [ "Closed weights with no dated snapshot identifiers or retirement policy", "No fixable snapshot identity beyond the launch-dated release" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/xai/grok-4.7.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://docs.x.ai/developers/models/grok-4.7", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Grok 4.7 announcement", "url": "https://x.ai/news/grok-4-7", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-xai-grok-imagine-image-2-0", "source_id": "xai/grok-imagine-image-2.0", "name": "Grok Imagine Image 2.0", "developer": "SpaceXAI (xAI)", "url": "https://docs.x.ai/developers/models/grok-imagine-image-2.0", "description": "xAI's Grok Imagine Image 2.0, a text-and-image-input image generation and editing model served through the Grok API Imagine endpoints with selectable resolution and quality, and the model xAI points to as its current image model.", "model_type": "image_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Grok Imagine Image 2.0", "description": "Image model for prompt-driven generation, editing, and visual design workflows", "family": "grok", "release_date": "2026-08-07", "last_updated": "2026-08-07", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 8000, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Grok Imagine Image 2.0 is served, not distributed, under the xAI enterprise terms and acceptable use policy that already govern the reviewed Grok records. The policy forbids using the service or its output to develop competing machine learning models and forbids distilling model data or outputs. No weights are released.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "scope": "Business access to Grok Imagine image generation through the SpaceXAI API and related services", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-xAI-AUP", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into xAI service access", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents Grok Imagine Image 2.0 under the API model name grok-imagine-image-2.0. The retiring grok-imagine-image-quality slug, which xAI will redirect to this model, the original grok-imagine-image 1.0, the Grok Imagine video models, and the Grok consumer apps are separate boundaries.", "strengths": [ "A dedicated model page names the slug, supported regions, batch availability, and per-resolution and per-quality pricing", "Release notes document behavior changes to the quality parameter and the successor role for the retiring Image Quality slug" ], "tradeoffs": [ "Closed weights served only through the xAI API under terms that bar competing-model development and distillation", "No dated snapshot identifier is documented, and the default quality moved from medium to auto in a release note" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/xai/grok-imagine-image-2.0.toml", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official model page for grok-imagine-image-2.0", "url": "https://docs.x.ai/developers/models/grok-imagine-image-2.0", "verified_at": "2026-10-06" }, { "kind": "web", "label": "xAI model overview naming Grok Imagine Image 2.0 as the image model", "url": "https://docs.x.ai/developers/models", "verified_at": "2026-10-06" }, { "kind": "web", "label": "xAI release notes documenting the quality parameter change and the Image Quality retirement", "url": "https://docs.x.ai/developers/release-notes", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-xai-grok-imagine-image-quality", "source_id": "xai/grok-imagine-image-quality", "name": "Grok Imagine Image Quality", "developer": "SpaceXAI (xAI)", "url": "https://docs.x.ai/developers/models/grok-imagine-image-quality", "description": "xAI's Grok Imagine Image Quality, a text-and-image-input image generation and editing model with the dated alias grok-imagine-image-quality-20260403, scheduled for retirement on November 2, 2026 when its slug will be redirected to Grok Imagine Image 2.0.", "model_type": "image_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Grok Imagine Image Quality", "description": "Higher-fidelity Grok Imagine image model for prompt-driven generation, editing, and visual design workflows", "family": "grok", "release_date": "2026-04-03", "last_updated": "2026-04-03", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 16000, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Grok Imagine Image Quality is served, not distributed, under the xAI enterprise terms and acceptable use policy that already govern the reviewed Grok records. The policy forbids using the service or its output to develop competing machine learning models and forbids distilling model data or outputs. No weights are released.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "scope": "Business access to Grok Imagine Image Quality through the SpaceXAI API and related services", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-xAI-AUP", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into xAI service access", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents Grok Imagine Image Quality under the API model name grok-imagine-image-quality and its dated alias grok-imagine-image-quality-20260403. The successor Grok Imagine Image 2.0, the original grok-imagine-image 1.0, the grok-imagine-image-pro alias route, and the Grok Imagine video models are separate boundaries.", "strengths": [ "A dedicated model page names a dated alias, the regions, and per-resolution pricing", "xAI published a 60-day retirement notice and a migration guide naming the redirect target and the price change" ], "tradeoffs": [ "The slug is retired on November 2, 2026 and afterward serves a different model at low quality, so the release has a terminating lifecycle", "Closed weights served only through the xAI API, with no batch support and no hosting by other operators" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 4, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.6 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/xai/grok-imagine-image-quality.toml", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official model page for grok-imagine-image-quality with its dated alias", "url": "https://docs.x.ai/developers/models/grok-imagine-image-quality", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Migration guide for the November 2, 2026 retirement and redirect", "url": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2", "verified_at": "2026-10-06" }, { "kind": "web", "label": "xAI release notes announcing the retirement on September 2, 2026", "url": "https://docs.x.ai/developers/release-notes", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-xai-grok-imagine-video-1-5", "source_id": "xai/grok-imagine-video-1.5", "name": "Grok Imagine Video 1.5", "developer": "SpaceXAI (xAI)", "url": "https://docs.x.ai/developers/models/grok-imagine-video-1.5", "description": "xAI's Grok Imagine Video 1.5, a video generation model with text, image, and audio-voice inputs for text-to-video, image-to-video, and reference-to-video at up to 1080p, served through the Grok API Imagine endpoints.", "model_type": "video_generation_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Grok Imagine Video 1.5", "description": "Video model for image-to-video generation, editing, and extension workflows", "family": "grok", "release_date": "2026-05-30", "last_updated": "2026-05-30", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "video" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 1024, "input": null, "output": 0 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Grok Imagine Video 1.5 is served, not distributed, under the xAI enterprise terms and acceptable use policy that already govern the reviewed Grok records. The policy forbids using the service or its output to develop competing machine learning models and forbids distilling model data or outputs. No weights are released.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "scope": "Business access to Grok Imagine video generation through the SpaceXAI API and related services", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-10-06" }, { "license_id": "LicenseRef-xAI-AUP", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into xAI service access", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents Grok Imagine Video 1.5 under the API model name grok-imagine-video-1.5 and its dated alias grok-imagine-video-1.5-2026-05-30. The grok-imagine-video-1.5-lite model, which has its own model page and pricing, the original grok-imagine-video, and the Grok Imagine image models are separate boundaries.", "strengths": [ "A dedicated model page names a dated alias, the preview alias, per-resolution pricing, rate limits, and batch availability", "Release notes document the July 31 change to the supported input modes and native 1080p" ], "tradeoffs": [ "Closed weights served only through the xAI API under terms that bar competing-model development and distillation", "The model page lists a single serving region and documents no retirement policy" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 3.86 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/xai/grok-imagine-video-1.5.toml", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official model page for grok-imagine-video-1.5 with its dated alias", "url": "https://docs.x.ai/developers/models/grok-imagine-video-1.5", "verified_at": "2026-10-06" }, { "kind": "web", "label": "xAI video generation guide naming grok-imagine-video-1.5 and its 1080p support", "url": "https://docs.x.ai/developers/model-capabilities/video/generation", "verified_at": "2026-10-06" }, { "kind": "web", "label": "xAI release notes documenting the July 31 modality change", "url": "https://docs.x.ai/developers/release-notes", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-xai-grok-voice-stt-1-0", "source_id": "xai/grok-voice-stt-1.0", "name": "Grok Voice Transcribe 1.0", "developer": "SpaceXAI (xAI)", "url": "https://docs.x.ai/developers/model-capabilities/audio/speech-to-text", "description": "xAI's Grok Voice Transcribe 1.0, a speech-to-text model released April 17, 2026, transcribing audio in 25 languages through the STT API and WebSocket endpoint, being superseded by the Transcribe 2.0 release.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Grok Voice STT 1.0", "description": "Grok Voice STT 1.0 is xAI's speech-to-text model. It supports transcription with word-level timestamps, optional speaker diarization, and multichannel audio.", "family": null, "release_date": "2026-08-04", "last_updated": "2026-08-04", "knowledge_cutoff": null, "modalities": { "input": [ "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 15000, "input": null, "output": 15000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Grok Voice Transcribe 1.0 is served, not distributed, under the xAI enterprise terms and acceptable use policy used by the reviewed Grok records. The models.dev row names the catalog-invented 'Grok Voice STT 1.0'; the first-party identity is grok-voice-transcribe-1.0.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "scope": "Access to the Grok Voice Transcribe 1.0 STT API and related xAI services", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-22" }, { "license_id": "LicenseRef-xAI-AUP", "scope": "Acceptable-use and anti-distillation restrictions incorporated into xAI service access", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Grok Voice Transcribe 1.0 speech-to-text release served through the xAI STT API, launched April 17, 2026, and named by the pinned grok-voice-transcribe-1.0 slug. The Transcribe 2.0 successor, the speech-to-speech voice lines, and the Grok 4.x lines are separate boundaries.", "strengths": [ "Dated first-party announcement and a dedicated docs page naming the model with endpoints, pricing, and limits", "License path identical to the already-verified Grok records, unambiguous for an API-only line" ], "tradeoffs": [ "The models.dev identity 'Grok Voice STT 1.0' diverges from the first-party grok-voice-transcribe-1.0 slug", "Already being superseded by Transcribe 2.0 and scheduled for deprecation, so the line carries a terminating lifecycle" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 4, "documentation_provenance": 7, "overall": 3.76 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/xai/grok-voice-stt-1.0.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://docs.x.ai/developers/model-capabilities/audio/speech-to-text", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Grok STT and TTS APIs announcement", "url": "https://x.ai/news/grok-stt-and-tts-apis", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-stepfun-step-3-5-flash", "source_id": "stepfun/step-3.5-flash", "name": "Step 3.5 Flash", "developer": "StepFun", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash", "description": "StepFun's 196-billion-parameter sparse MoE flagship reasoning release with 11B active parameters, published as Apache-declared open weights with a managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Step 3.5 Flash", "description": "StepFun flash lane for quick multimodal reasoning and coding assistance", "family": null, "release_date": "2026-01-29", "last_updated": "2026-02-13", "knowledge_cutoff": "2025-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": 256000, "output": 256000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the model card frontmatter, badge, and footer declare the Apache 2.0 license with no use restrictions attached. The grant rests on those concordant first-party declarations pointing at the standard Apache-2.0 text.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Step-3.5-Flash weights as declared on the official model card, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the February 2026 Step-3.5-Flash release at Hugging Face commit ab446a3de5e171ea341227e24bb1f090e1b771f7 (last modified 2026-03-17), together with the step-3.5-flash API model. The 2603 agent-optimized variant, the 3.7 successor, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 196B MoE checkpoint with Base, FP8, and GGUF siblings under a declared Apache-2.0 grant", "First-party API with 256K context plus vLLM, SGLang, and Transformers recipes" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card declarations", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 9, "overall": 7.92 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint declaring the Apache-2.0 grant", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official StepFun step-3.5-flash API documentation", "url": "https://platform.stepfun.ai/docs/en/guides/models/step-3.5-flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/stepfun/step-3.5-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-stepfun-step-3-7-flash", "source_id": "stepfun/step-3.7-flash", "name": "Step 3.7 Flash", "developer": "StepFun", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash", "description": "StepFun's 198-billion-parameter multimodal flagship with a vision encoder for native image and video understanding, published as Apache-declared open weights with a managed API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Step 3.7 Flash", "description": "Newer StepFun flash model for faster agents, coding, and multimodal prompts", "family": null, "release_date": "2026-05-29", "last_updated": "2026-05-29", "knowledge_cutoff": "2026-03-01", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": 256000, "output": 256000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "No LICENSE file ships in the weight repository; the model card frontmatter, badge, and footer declare the Apache 2.0 license with no use restrictions attached. The grant rests on those concordant first-party declarations pointing at the standard Apache-2.0 text.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Step-3.7-Flash weights as declared on the official model card, pointing at the standard Apache-2.0 text", "kind": "web_terms", "url": "https://www.apache.org/licenses/LICENSE-2.0", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the May 2026 Step-3.7-Flash release at Hugging Face commit 5f6244077ac62e04eec3f320501ff8c2b293373a (last modified 2026-06-03), a distinct multimodal line rather than a 3.5 snapshot, together with the step-3.7-flash API model. The 3.5 generation, later StepFun lines, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 198B multimodal checkpoint with FP8, NVFP4, and GGUF variants under a declared Apache-2.0 grant", "First-party API with native image and video understanding plus three reasoning effort levels" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on card declarations", "Version identity rests on repository commits with no stated retirement posture" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 9, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 9, "overall": 7.92 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint declaring the Apache-2.0 grant", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official StepFun step-3.7-flash API documentation", "url": "https://platform.stepfun.ai/docs/en/guides/models/step-3.7-flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/stepfun/step-3.7-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-stepfun-step-5-preview", "source_id": "stepfun/step-5-preview", "name": "Step 5 Preview", "developer": "StepFun", "url": "https://www.stepfun.com/step-5-preview", "description": "StepFun's Step 5 Preview, a reasoning-focused sparse-MoE model with 600B total and 27B active parameters, 1M context, and vision input, announced September 20, 2026, with weights scheduled to open on October 15, 2026.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Step 5 Preview", "description": "StepFun's next-generation flagship base model for coding and professional knowledge work, with native text, image, and video input and a 1M-token context window", "family": null, "release_date": "2026-09-16", "last_updated": "2026-09-20", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 1000000, "input": 1000000, "output": 1000000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Step 5 Preview is currently served through the StepFun Open Platform API with closed weights; the governing terms are the StepFun Open Platform service terms. The announcement schedules weight release for October 15, 2026, which is not yet distributable.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "Step 5 Preview served through the StepFun Open Platform API under StepFun service terms", "kind": "web_terms", "url": "https://platform.stepfun.ai/docs/en/guides/models/step-5-preview", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Step 5 Preview release announced September 20, 2026, a distinct line from the reviewed Step 3.5 Flash and Step 3.7 Flash records. The scheduled weight release and the earlier Step 3.x lines are separate boundaries.", "strengths": [ "Clear dated first-party announcement and platform docs establishing the Step 5 line identity", "Distinct 1M-context vision-capable reasoning architecture with reasoning_effort support" ], "tradeoffs": [ "Closed weights until the scheduled October 15, 2026 release, so the record is API-only today", "Preview label and a very fresh release with a thin ecosystem runtime" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.14 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/stepfun/step-5-preview.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://www.stepfun.com/step-5-preview", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-swiss-ai-apertus-70b", "source_id": "swiss-ai/apertus-70b", "name": "Apertus 70B", "developer": "Swiss AI Initiative", "url": "https://huggingface.co/swiss-ai/Apertus-70B-2509", "description": "A 70B multilingual text model pretrained on 15T tokens of open data, released under Apache-2.0 with a 65,536-token context and an advisory usage policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Apertus 70B", "description": "Fully open 70B multilingual LLM supporting 1800+ languages with 65K context. Trained on 15T tokens of compliant open data. Apache 2.0, EU AI Act compliant.", "family": null, "release_date": "2025-09-02", "last_updated": "2025-09-02", "knowledge_cutoff": "2025-09", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [ { "label": "Paper", "url": "https://arxiv.org/abs/2509.14233", "type": "paper" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships a LICENSE.txt with the standard Apache License 2.0 text (\"Copyright 2025 - The Swiss AI team.\"), confirmed at the pinned commit, and the lab states the weights are \"released under the Apache 2.0 licence, which allows commercial use.\" A separate advisory USAGE_POLICY.md adds indemnity, data-protection, and output-filter duties but no use restriction.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Apertus-70B-2509 weights, tokenizer, and configuration in the official swiss-ai repository", "kind": "web_terms", "url": "https://huggingface.co/swiss-ai/Apertus-70B-2509/blob/379311a08b6e691f7b5cfbc1c408e8fc0c172981/LICENSE.txt", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the swiss-ai/Apertus-70B-2509 base release of September 2025 at Hugging Face commit 379311a08b6e691f7b5cfbc1c408e8fc0c172981. The 8B sibling, the Instruct fine-tune, the 1.5 generation, quantizations, and community fine-tunes remain separate boundaries.", "strengths": [ "Ungated Apache-2.0 weights with the full license text at the pinned commit", "Dated release identity with a model card, tech report, and 65,536-token context documentation" ], "tradeoffs": [ "A separate acceptable-use policy adds indemnity, personal-data, and output-filter duties", "No inference provider or first-party API serves this release, so reach is self-host only" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 10, "deployment_portability": 9, "serving_reach": 3, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.22 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/swiss-ai/Apertus-70B-2509", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Apache-2.0 license text at pinned commit", "url": "https://huggingface.co/swiss-ai/Apertus-70B-2509/blob/379311a08b6e691f7b5cfbc1c408e8fc0c172981/LICENSE.txt", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/swiss-ai/apertus-70b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-swiss-ai-apertus-8b", "source_id": "swiss-ai/apertus-8b", "name": "Apertus 8B", "developer": "Swiss AI Initiative", "url": "https://huggingface.co/swiss-ai/Apertus-8B-2509", "description": "An 8B multilingual text model pretrained on 15T tokens of open data, released under Apache-2.0 with a 65,536-token context and an advisory usage policy.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Apertus 8B", "description": "Fully open 8B multilingual LLM supporting 1800+ languages with 65K context. Trained on compliant open data. Apache 2.0, EU AI Act compliant.", "family": null, "release_date": "2025-09-02", "last_updated": "2025-09-02", "knowledge_cutoff": "2025-09", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": false, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 8192 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [ { "label": "Paper", "url": "https://arxiv.org/abs/2509.14233", "type": "paper" } ], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships a LICENSE.txt with the standard Apache License 2.0 text (\"Copyright 2025 - The Swiss AI team.\"), confirmed at the pinned commit, and the lab states the weights are \"released under the Apache 2.0 licence, which allows commercial use.\" A separate advisory USAGE_POLICY.md adds indemnity, data-protection, and output-filter duties but no use restriction.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Apertus-8B-2509 weights, tokenizer, and configuration in the official swiss-ai repository", "kind": "web_terms", "url": "https://huggingface.co/swiss-ai/Apertus-8B-2509/blob/3162c99675aa588097cecd4a24b9aa1f712af477/LICENSE.txt", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the swiss-ai/Apertus-8B-2509 base release of September 2025 at Hugging Face commit 3162c99675aa588097cecd4a24b9aa1f712af477. The 70B sibling, the Instruct fine-tune, the 1.5 generation, quantizations, and community fine-tunes remain separate boundaries.", "strengths": [ "Ungated Apache-2.0 weights with the full license text at the pinned commit", "Dated release identity with a model card, tech report, and 65,536-token context documentation" ], "tradeoffs": [ "A separate acceptable-use policy adds indemnity, personal-data, and output-filter duties", "No inference provider or first-party API serves this release, so reach is self-host only" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 10, "deployment_portability": 9, "serving_reach": 3, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.22 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoints", "url": "https://huggingface.co/swiss-ai/Apertus-8B-2509", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Apache-2.0 license text at pinned commit", "url": "https://huggingface.co/swiss-ai/Apertus-8B-2509/blob/3162c99675aa588097cecd4a24b9aa1f712af477/LICENSE.txt", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/swiss-ai/apertus-8b.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-tencent-hunyuan-a13b-instruct", "source_id": null, "name": "Hunyuan-A13B-Instruct", "developer": "Tencent", "url": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct", "description": "Tencent's 80-billion-parameter Mixture-of-Experts instruct release with 13B active parameters and switchable fast and slow thinking, published as downloadable weights under a territory-limited community license.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Hunyuan-A13B-Instruct", "description": "Tencent Hunyuan fine-grained MoE instruct model with hybrid fast and slow reasoning and a 256K context window", "family": "Hunyuan", "release_date": "2025-06-27", "last_updated": "2025-08-21", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct" } ] }, "licenses": [ "LicenseRef-Tencent-Hunyuan-Community-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository LICENSE file is the Tencent Hunyuan Community License Agreement for the June 27, 2025 A13B release. It does not apply in the European Union, United Kingdom, or South Korea and licenses use only inside its Territory. It bars using the works or their Output to improve any other AI model, prohibits military use through its Acceptable Use Policy, and requires a separate Tencent license above 100 million monthly active users, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Tencent-Hunyuan-Community-License", "scope": "Tencent Hunyuan A13B weights, model code, and documentation including Model Derivatives, within the Territory only", "kind": "web_terms", "url": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct/blob/290ddb9a56ed23c2c83a1c8081533e58925df952/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the Hunyuan-A13B-Instruct weight release of June 2025 at Hugging Face commit 290ddb9a56ed23c2c83a1c8081533e58925df952 (last modified 2025-08-21). The Hunyuan-A13B-Pretrain base, the FP8 and GPTQ-Int4 quantized repackagings, later Tencent Hy generations, and any Tencent Cloud API listing remain separate boundaries.", "strengths": [ "Ungated 80B MoE checkpoint with the full custom license text and a Notice file beside the weights", "Model card documents TensorRT-LLM, vLLM, and SGLang serving with prebuilt Docker images and a 256K context recipe" ], "tradeoffs": [ "License excludes the European Union, United Kingdom, and South Korea and forbids using outputs to improve other AI models", "Model card names no hosted endpoint for this exact release, and the default config caps context at 32K until edited" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.24 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and deployment guidance", "url": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Tencent-Hunyuan GitHub repository and LICENSE", "url": "https://github.com/Tencent-Hunyuan/Hunyuan-A13B", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-tencent-hy-mt2-1-8b", "source_id": null, "name": "Hy-MT2-1.8B", "developer": "Tencent", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B", "description": "Tencent's 1.8-billion-parameter dense fast-thinking translation model from the Hy-MT2 family, covering 33 languages and published as downloadable Apache-2.0 weights with a training pipeline.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Hy-MT2-1.8B", "description": "Tencent Hy multilingual translation model for 33 languages with translation-instruction following", "family": "Hy-MT", "release_date": "2026-05-21", "last_updated": "2026-05-26", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.txt states that Hy-MT2-1.8B is licensed under the Apache License, Version 2.0, followed by the unmodified Apache 2.0 terms, with no territory limit, usage threshold, attribution gate, or acceptable-use policy added.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Hy-MT2-1.8B weights, tokenizer, configuration, and training files in the tencent/Hy-MT2-1.8B repository", "kind": "web_terms", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B/blob/9a341cd1b679d3efd23b46e847b01745a71ed792/LICENSE.txt", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the tencent/Hy-MT2-1.8B weight release announced on 2026-05-21 at Hugging Face commit 9a341cd1b679d3efd23b46e847b01745a71ed792 (last modified 2026-05-26). The other Hy-MT2 sizes, the FP8, GGUF, and low-bit quantized repackagings, the earlier HY-MT1.5 and Hunyuan-MT releases, and any hosted translation service remain separate boundaries.", "strengths": [ "Ungated checkpoint with a dedicated Apache-2.0 LICENSE.txt, full LoRA and full fine-tuning scripts, and vLLM and SGLang recipes in the repository", "Model card documents recommended sampling settings and a per-language prompt format for translation" ], "tradeoffs": [ "Model card names no hosted endpoint for this exact release, so served access must be arranged independently", "Loading needs transformers 5.6 or newer with trust_remote_code, and version identity rests on a moving main branch" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.0 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and deployment guidance", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Tencent-Hunyuan Hy-MT2 GitHub repository", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-tencent-hy-mt2-30b-a3b", "source_id": null, "name": "Hy-MT2-30B-A3B", "developer": "Tencent", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B", "description": "Tencent's 30-billion-parameter Mixture-of-Experts (3B active) fast-thinking translation model from the Hy-MT2 family, covering 33 languages and published as downloadable Apache-2.0 weights with a training pipeline.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Hy-MT2-30B-A3B", "description": "Tencent Hy multilingual translation model for 33 languages with translation-instruction following", "family": "Hy-MT", "release_date": "2026-05-21", "last_updated": "2026-07-21", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.txt states that Hy-MT2-30B-A3B is licensed under the Apache License, Version 2.0, followed by the unmodified Apache 2.0 terms, with no territory limit, usage threshold, attribution gate, or acceptable-use policy added.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Hy-MT2-30B-A3B weights, tokenizer, configuration, and training files in the tencent/Hy-MT2-30B-A3B repository", "kind": "web_terms", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/LICENSE.txt", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the tencent/Hy-MT2-30B-A3B weight release announced on 2026-05-21 at Hugging Face commit d3ead4dba61c09aac60a261a96ad1df3e705febb (last modified 2026-07-21). The other Hy-MT2 sizes, the FP8, GGUF, and low-bit quantized repackagings, the earlier HY-MT1.5 and Hunyuan-MT releases, and any hosted translation service remain separate boundaries.", "strengths": [ "Ungated checkpoint with a dedicated Apache-2.0 LICENSE.txt, full LoRA and full fine-tuning scripts, and vLLM and SGLang recipes in the repository", "Model card documents recommended sampling settings and a per-language prompt format for translation" ], "tradeoffs": [ "Model card names no hosted endpoint for this exact release, so served access must be arranged independently", "Loading needs transformers 5.6 or newer with trust_remote_code, and version identity rests on a moving main branch" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.0 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and deployment guidance", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Tencent-Hunyuan Hy-MT2 GitHub repository", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-tencent-hy-mt2-7b", "source_id": null, "name": "Hy-MT2-7B", "developer": "Tencent", "url": "https://huggingface.co/tencent/Hy-MT2-7B", "description": "Tencent's 7-billion-parameter dense fast-thinking translation model from the Hy-MT2 family, covering 33 languages and published as downloadable Apache-2.0 weights with a training pipeline.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Hy-MT2-7B", "description": "Tencent Hy multilingual translation model for 33 languages with translation-instruction following", "family": "Hy-MT", "release_date": "2026-05-21", "last_updated": "2026-05-26", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 262144, "input": null, "output": null }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hy-MT2-7B" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE.txt states that Hy-MT2-7B is licensed under the Apache License, Version 2.0, followed by the unmodified Apache 2.0 terms, with no territory limit, usage threshold, attribution gate, or acceptable-use policy added.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Hy-MT2-7B weights, tokenizer, configuration, and training files in the tencent/Hy-MT2-7B repository", "kind": "web_terms", "url": "https://huggingface.co/tencent/Hy-MT2-7B/blob/9b0eb4e8f001def3e5ff6469a0ac96fdb39ec223/LICENSE.txt", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the tencent/Hy-MT2-7B weight release announced on 2026-05-21 at Hugging Face commit 9b0eb4e8f001def3e5ff6469a0ac96fdb39ec223 (last modified 2026-05-26). The other Hy-MT2 sizes, the FP8, GGUF, and low-bit quantized repackagings, the earlier HY-MT1.5 and Hunyuan-MT releases, and any hosted translation service remain separate boundaries.", "strengths": [ "Ungated checkpoint with a dedicated Apache-2.0 LICENSE.txt, full LoRA and full fine-tuning scripts, and vLLM and SGLang recipes in the repository", "Model card documents recommended sampling settings and a per-language prompt format for translation" ], "tradeoffs": [ "Model card names no hosted endpoint for this exact release, so served access must be arranged independently", "Loading needs transformers 5.6 or newer with trust_remote_code, and version identity rests on a moving main branch" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.0 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and deployment guidance", "url": "https://huggingface.co/tencent/Hy-MT2-7B", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Tencent-Hunyuan Hy-MT2 GitHub repository", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" }, { "id": "model-tencent-hy3", "source_id": "tencent/hy3", "name": "Hy3", "developer": "Tencent", "url": "https://huggingface.co/tencent/Hy3", "description": "Tencent's 295-billion-parameter Mixture-of-Experts release with 21B active parameters, published as Apache-declared open weights with TokenHub API serving.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Hy3", "description": "Tencent Hy reasoning model for coding, instruction following, and agent tasks", "family": "Hy", "release_date": "2026-07-06", "last_updated": "2026-07-06", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": 192000, "output": 128000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hy3" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file declares the release licensed under Apache-2.0, matching the model card, HF tag, and July 2026 launch press release. No use restrictions accompany the declaration, so the release is open source.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Tencent Hy3 weights as declared in the repository LICENSE file and launch press release", "kind": "web_terms", "url": "https://huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the July 2026 Tencent Hy3 release at Hugging Face commit a960ebc3da325ba167f069f76c41eb62c9280d22 (last modified 2026-07-20), together with the hy3 TokenHub API model. The Hy3-preview predecessor, the Hy4 generation, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 295B MoE checkpoint under a declared Apache-2.0 grant with press corroboration", "First-party TokenHub serving under the exact hy3 model ID with 256K context" ], "tradeoffs": [ "The license declaration is short-form rather than a full license text in-repo", "Version identity rests on a moving main branch with no immutable tag" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 8, "documentation_provenance": 9, "overall": 8.32 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/tencent/Hy3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Tencent Hy3 launch press release (July 2026)", "url": "https://www.tencent.com/tencent-hunyuan-officially-releases-hy3-advancing-agent-capabilities-and-deeper-product-integration/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/tencent/hy3.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-tencent-hy3-preview", "source_id": "tencent/hy3-preview", "name": "Hy3 preview", "developer": "Tencent", "url": "https://huggingface.co/tencent/Hy3-preview", "description": "Tencent's 295-billion-parameter preview release, published as open weights under the geo-fenced Tencent Hy Community License with a retired managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Hy3 preview", "description": "Tencent Hy reasoning model for coding, instruction following, and agent tasks", "family": "Hy", "release_date": "2026-04-20", "last_updated": "2026-04-20", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 64000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hy3-preview" } ] }, "licenses": [ "LicenseRef-Tencent-Hy-Community-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The repository LICENSE file is the Tencent Hy Community License Agreement dated April 23, 2026, which excludes the European Union, United Kingdom, and South Korea from its territory and gates commercial use above 100 million monthly active users behind a separate license, so the release is source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Tencent-Hy-Community-License", "scope": "Tencent Hy3 preview weights under the in-repository community license", "kind": "web_terms", "url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the April 2026 Tencent Hy3 preview release at Hugging Face commit 549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a, frozen since launch. Its managed API retired with the line superseded by Hy3. The Hy3 successor, the Hy4 generation, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 295B preview checkpoint with its full custom license text in-repo", "Frozen since April 2026 with a clean supersession line to Hy3" ], "tradeoffs": [ "Territory exclusion plus a commercial-use gate make this source-available, not open", "The managed API retired, leaving self-operation as the supported path" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 4, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 6.66 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/tencent/Hy3-preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Tencent Hy3 preview launch press release (April 2026)", "url": "https://www.tencent.com/tencent-unveils-hy3-preview-model-enhances-agent-capabilities-and-real-world-usability/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/tencent/hy3-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-tencent-hy4-preview", "source_id": "tencent/hy4-preview", "name": "Hy4 preview", "developer": "Tencent", "url": "https://huggingface.co/tencent/Hy4-preview", "description": "Tencent's 770-billion-parameter next-generation preview with 49B active parameters and 1M context, published as Apache-declared open weights with TokenHub API serving.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "Hy4 preview", "description": "A next-generation productivity model with significantly enhanced Agent and complex task execution capabilities.", "family": "Hy", "release_date": "2026-08-28", "last_updated": "2026-08-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1024000, "input": null, "output": 64000 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository LICENSE file declares the release licensed under Apache-2.0, matching the model card and August 2026 launch press release. No use restrictions accompany the declaration, so the release is open source.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Tencent Hy4 preview weights as declared in the repository LICENSE file and launch press release", "kind": "web_terms", "url": "https://huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/LICENSE", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the August 2026 Tencent Hy4 preview release at Hugging Face commit 705d81ee51566a186d645b74c974d642ef2828fe (last modified 2026-08-28), a new 770B/49B architecture rather than a Hy3 snapshot, together with the hy4-preview TokenHub API model. Earlier Hy generations and community quantizations remain separate boundaries.", "strengths": [ "Ungated 770B next-generation checkpoint under a declared Apache-2.0 grant", "First-party TokenHub serving under the exact hy4-preview model ID with 1M context" ], "tradeoffs": [ "Self-declared early version with fast-moving commits in its first days", "770B total parameters need cluster-class serving for self-operation" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.22 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/tencent/Hy4-preview", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Tencent Hy4 preview launch press release (August 2026)", "url": "https://www.tencent.com/tencent-releases-and-open-sources-tencent-hy4-preview/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/tencent/hy4-preview.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-thinkingmachines-inkling", "source_id": "thinkingmachines/inkling", "name": "Inkling", "developer": "Thinking Machines Lab", "url": "https://huggingface.co/thinkingmachines/Inkling", "description": "Thinking Machines Lab's Inkling, a roughly 1-trillion-parameter multimodal MoE with 41B active parameters and 1M context accepting text, image, and audio input, published as open weights under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Inkling", "description": "Multimodal MoE reasoning model (975B total, 41B active) for text, image, and audio", "family": "ling", "release_date": "2026-07-15", "last_updated": "2026-07-15", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 1048576 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/thinkingmachines/Inkling" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no separate LICENSE file; the pinned model-card metadata declares Apache-2.0 with an explicit license link to the canonical Apache-2.0 text, and the NVFP4 variant declares the same grant.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Inkling repository weights, Safetensors checkpoints, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/thinkingmachines/Inkling", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Inkling weight release announced July 15, 2026, a multimodal MoE accepting text, image, and audio. The Inkling-Small release, quantized variants, and independent hosts are separate boundaries.", "strengths": [ "Ungated open multimodal weights at frontier scale with a detailed first-party card, evaluation, and safety material", "Day-zero vLLM, SGLang, and transformers support with BF16 and NVFP4 variants" ], "tradeoffs": [ "The repository ships no LICENSE file; Apache-2.0 rests on pinned model-card metadata", "BF16 serving needs roughly 2TB of VRAM across multiple nodes, and no dated deprecation or lifecycle statement exists" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 5, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.16 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/thinkingmachines/inkling.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/thinkingmachines/Inkling", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-thinkingmachines-inkling-small", "source_id": "thinkingmachines/inkling-small", "name": "Inkling-Small", "developer": "Thinking Machines Lab", "url": "https://huggingface.co/thinkingmachines/Inkling-Small", "description": "Thinking Machines Lab's Inkling-Small, the smaller sibling of the Inkling family with 276B total and 12B active parameters, natively multimodal with 1M context, published as open weights under Apache-2.0.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Inkling Small", "description": "Multimodal MoE reasoning model (276B total, 12B active) for text, image, and audio", "family": "ling", "release_date": "2026-07-30", "last_updated": "2026-07-30", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 1048576 }, "reported_open_weights": true, "reported_license": "Apache-2.0", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/thinkingmachines/Inkling-Small" } ] }, "licenses": [ "Apache-2.0" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no separate LICENSE file; the pinned model-card metadata declares Apache-2.0 with an explicit license link to the canonical Apache-2.0 text, matching the family grant.", "license_evidence": [ { "license_id": "Apache-2.0", "scope": "Inkling-Small repository weights, Safetensors checkpoints, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/thinkingmachines/Inkling-Small", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Inkling-Small weight release, the smaller sibling of the Inkling family. The larger Inkling release, quantized variants, and independent hosts are separate boundaries.", "strengths": [ "Distinct smaller sibling that remains natively multimodal with a practical VRAM envelope for self-hosting", "Ungated weights with one-click inference endpoints and strong evaluation against similar-scale peers" ], "tradeoffs": [ "The repository ships no LICENSE file; Apache-2.0 rests on pinned model-card metadata", "Audio input performance is only self-reported on a limited benchmark set, and no lifecycle statement exists" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.36 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/thinkingmachines/inkling-small.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/thinkingmachines/Inkling-Small", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-tiiuae-falcon-h1-7b-instruct", "source_id": null, "name": "Falcon-H1 7B Instruct", "developer": "TII", "url": "https://huggingface.co/tiiuae/Falcon-H1-7B-Instruct", "description": "TII's May 2025 Falcon-H1 instruction release at 7B parameters, a performance-and-portability model unveiled in Abu Dhabi alongside Falcon Arabic, published as open weights under the Apache-based Falcon LLM License with its acceptable-use policy. No developer API serves this checkpoint. Reviewed by automation under ADR 055 with human spot-checks; verify against the cited evidence.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Falcon-H1 7B Instruct", "description": "7B instruction release of the Falcon-H1 high-performance portable model family", "family": "falcon", "release_date": "2025-05-21", "last_updated": "2025-05-21", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": true }, "limits": { "context": null, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Falcon-LLM", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/tiiuae/Falcon-H1-7B-Instruct" } ] }, "licenses": [ "LicenseRef-Falcon-LLM-License" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The tiiuae/Falcon-H1-7B-Instruct model card names the Falcon-LLM License and links TII's Falcon terms page, which states the licence is based on Apache License 2.0 with modifications and incorporates an Acceptable Use Policy, hosted on FalconLLM.tii.ae, that users and downstream distributors must follow. The commercial grant plus use-policy duties make the release source-available rather than open source.", "license_evidence": [ { "license_id": "LicenseRef-Falcon-LLM-License", "scope": "Falcon-H1-7B-Instruct weights published in TII's Hugging Face repository under the Falcon LLM License named by the model card", "kind": "web_terms", "url": "https://falconllm.tii.ae/falcon-terms-and-conditions.html", "verified_at": "2026-10-13" } ], "access_boundary": "This record represents the Falcon-H1-7B-Instruct checkpoint of the Falcon-H1 family. The base model, the 0.5B through 34B siblings, the Falcon 3 and Falcon 2 generations, the Falcon Arabic line, and the Falcon-H1R reasoning releases are separate boundaries.", "strengths": [ "Small-to-medium instruction checkpoint TII positions against larger LLaMA and Qwen offerings for on-device and resource-limited use", "Commercially usable Apache-based grant with the use policy published alongside the licence text on TII's own Falcon site" ], "tradeoffs": [ "The incorporated acceptable-use policy and licence modifications need legal review before regulated or client-facing use", "No developer API or managed host serves this checkpoint, so running it is entirely self-operated" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 7, "deployment_portability": 8, "serving_reach": 2, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.08 }, "evidence": [ { "kind": "web", "label": "TII Falcon-H1 and Falcon Arabic launch announcement: May 21, 2025, Abu Dhabi", "url": "https://www.tii.ae/news/middle-easts-leading-ai-powerhouse-tii-launches-two-new-ai-models-falcon-arabic-first-arabic", "verified_at": "2026-10-13" }, { "kind": "web", "label": "TII Falcon-H1-7B-Instruct weight repository: Falcon-LLM licence", "url": "https://huggingface.co/tiiuae/Falcon-H1-7B-Instruct", "verified_at": "2026-10-13" }, { "kind": "web", "label": "TII Falcon terms and conditions: Apache-based licence with acceptable-use duties", "url": "https://falconllm.tii.ae/falcon-terms-and-conditions.html", "verified_at": "2026-10-13" } ], "flags": [ { "kind": "maker_risk_safeguards", "status": "no_statement_found", "url": "https://huggingface.co/tiiuae/Falcon-H1-7B-Instruct", "verified_at": "2026-10-13", "research_confidence": "high" } ], "metadata_verified_at": "2026-10-13", "verified_at": "2026-10-13" }, { "id": "model-trendyol-asure-12b", "source_id": "trendyol/asure-12b", "name": "Trendyol-LLM-Asure-12B", "developer": "Trendyol", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B", "description": "Trendyol's Asure-12B, a Gemma-3-12B-based multimodal instruct model tuned for Turkish and English e-commerce, published as ungated open weights under the Gemma terms of use.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "third_party_hosting" ], "source_metadata": { "name": "Trendyol Asure 12B", "description": "Turkish-language multimodal instruct model built on Gemma 3 12B for e-commerce text, chat, and image-text tasks", "family": "gemma", "release_date": "2026-02-19", "last_updated": "2026-02-20", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "Gemma", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B" } ] }, "licenses": [ "LicenseRef-Gemma-Terms-of-Use" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The model card tags the Gemma license and the release is built on Gemma 3-12B, so the governing terms for the weights are the Gemma terms of use. The repository ships no separate LICENSE file, so the grant rests on the pinned model-card metadata and the Gemma base-model terms.", "license_evidence": [ { "license_id": "LicenseRef-Gemma-Terms-of-Use", "scope": "Trendyol-LLM-Asure-12B weights built on the Gemma 3-12B base under Gemma terms of use", "kind": "web_terms", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Trendyol-LLM-Asure-12B weight release, a Gemma-3-12B derivative instruct model tuned for Turkish and English e-commerce. The Gemma base models and independent hosts are separate boundaries.", "strengths": [ "Distinct named release with ungated multimodal open weights and documented e-commerce tuning", "Serving available through an independent host as well as direct weights" ], "tradeoffs": [ "The license grant is metadata-only with no shipped LICENSE file", "A niche e-commerce derivative of the Gemma 3-12B base with a limited general knowledge footprint" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 4, "lifecycle_transparency": 5, "documentation_provenance": 5, "overall": 5.56 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/trendyol/asure-12b.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-typesafe-jev-1-13-0", "source_id": null, "name": "Jev 1.13", "developer": "TypeSafe AI", "url": "https://docs.typesafe.ai/models", "description": "A closed text-input model served through TypeSafe's own API under the pinned ID jev-1.13.0 and listed on Vercel AI Gateway; it answers typed questions about supplied text with a selected option, a score, or a yes-probability instead of free-form prose, and no weights are published. The Atlas's own Hacker News signals routine calls this model, so this review rests on published documentation and terms only.", "model_type": "judgment_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Jev 1.13", "description": null, "family": null, "release_date": null, "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": null, "tool_call": null, "structured_output": null, "temperature": null }, "limits": { "context": 64000, "input": null, "output": null }, "reported_open_weights": null, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-TypeSafe-MCA" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "TypeSafe provides the release as a managed service under its Master Customer Agreement, last updated 23 September 2026, which a customer accepts through an order, TypeSafe's website checkout, or a confirmation email. The agreement prohibits offering the Services as a standalone service, using the Services or Output for model distillation, to train a model that imitates the output, or to build a similar or competing product, and reverse engineering the Services; it licenses no model weights for self-operation. It incorporates an Acceptable Use Policy, which resolved on 25 September 2026.", "license_evidence": [ { "license_id": "LicenseRef-TypeSafe-MCA", "kind": "web_terms", "url": "https://typesafe.ai/legal/mca", "verified_at": "2026-09-25", "scope": "Commercial access to the release through TypeSafe API keys and the System One endpoint, including the use restrictions on the Services and their Output" } ], "access_boundary": "This record represents the Jev 1.13 release identified by the versioned model ID jev-1.13.0, launched with TypeSafe's System One announcement of 15 September 2026. The moving aliases jev-latest and jev-preview, the System One API endpoint that serves it, Vercel AI Gateway and other gateways that route to it, TypeSafe's client SDKs and skills repository, and later Jev versions remain separate service, product, or release boundaries.", "strengths": [ "TypeSafe documents a versioned model ID, states that the jev-latest and jev-preview aliases move between releases, and reports the answering version in every response", "First-party documentation covers input limits, rate limits, language support, the training approach, and a dated page of known limitations for this version" ], "tradeoffs": [ "No developer-published weights are available for self-operation, and the documented gateway names the Jev line rather than the pinned version", "No deprecation policy or retirement commitment is published for the versioned ID, and rate limits are documented as subject to change without notice" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.66 }, "evidence": [ { "kind": "web", "label": "Official TypeSafe models page with the Jev 1.13 model ID, aliases, and limits", "url": "https://docs.typesafe.ai/models", "verified_at": "2026-09-20" }, { "kind": "web", "label": "Official Jev 1.13 known-limitations page", "url": "https://docs.typesafe.ai/model-jaggedness/jev-1.13", "verified_at": "2026-09-20" }, { "kind": "web", "label": "Official TypeSafe primer on the model's training approach and outputs", "url": "https://docs.typesafe.ai/introduction/machine-learning-primer", "verified_at": "2026-09-20" }, { "kind": "web", "label": "Official TypeSafe API reference for the System One endpoint", "url": "https://docs.typesafe.ai/api", "verified_at": "2026-09-20" }, { "kind": "web", "label": "Official TypeSafe announcement introducing System One models and Jev", "url": "https://typesafe.ai/blog/introducing-system-one-models-and-jev", "verified_at": "2026-09-20" }, { "kind": "web", "label": "Vercel AI Gateway model page for Jev", "url": "https://vercel.com/ai-gateway/models/jev", "verified_at": "2026-09-20" } ], "metadata_verified_at": "2026-09-20", "verified_at": "2026-09-20" }, { "id": "model-upstage-solar-mini4", "source_id": null, "name": "Solar Mini 4", "developer": "Upstage", "url": "https://console.upstage.ai/docs/models/solar-mini-4", "description": "Upstage's September 2026 Solar Mini 4 compact release for agentic use, with 35B total and 3B active parameters and 512K context, served as a managed API with no public artifact of any kind.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Solar Mini 4", "description": "Upstage's cost-efficient, compact model for agentic use cases where response speed and cost matter", "family": "solar-mini", "release_date": "2026-09-22", "last_updated": "2026-09-22", "knowledge_cutoff": "2026-02", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": null }, "limits": { "context": 524288, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No license of any kind is published for this closed commercial line: no weights, no tokenizer, no model terms beyond the Upstage Terms of Use. The release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of solar-mini4 with no published license", "kind": "web_terms", "url": "https://www.upstage.ai/terms-of-service", "verified_at": "2026-09-28" } ], "access_boundary": "This record represents the Solar Mini 4 release of September 22, 2026, served as the solar-mini4-260922 API version behind the solar-mini4 alias with 512K context and 128K max output. The Solar Jev decision model that Upstage serves on Solar Mini 4, earlier and larger Solar lines, and the platform itself are separate boundaries.", "strengths": [ "Dated solar-mini4-260922 version behind the solar-mini4 alias, documented with its release date, specs, and Console, Playground, and on-premises availability", "512K context with reasoning, structured outputs, and tool calling at 3B active parameters" ], "tradeoffs": [ "No downloadable artifact of any kind exists first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely", "Documented only by its Console model page and changelog entry, with no launch announcement" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 1, "deployment_portability": 5, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 7, "overall": 4.94 }, "evidence": [ { "kind": "web", "label": "Official Solar Mini 4 docs with version, specs, and serving", "url": "https://console.upstage.ai/docs/models/solar-mini-4", "verified_at": "2026-09-28" }, { "kind": "web", "label": "Upstage Console changelog entry for the Solar Mini 4 release on September 22, 2026", "url": "https://console.upstage.ai/docs/resources/changelog", "verified_at": "2026-09-28" } ], "metadata_verified_at": "2026-09-28", "verified_at": "2026-09-28" }, { "id": "model-upstage-solar-open2-250b", "source_id": null, "name": "Solar Open 2", "developer": "Upstage", "url": "https://huggingface.co/upstage/Solar-Open2-250B", "description": "Upstage's open-weight 250B-parameter mixture-of-experts language model with 15B active parameters, a 1M-token context from hybrid linear and softmax attention, and reasoning and tool calling for agentic work in Korean, English, and Japanese, published under the Apache-2.0-based Upstage Solar License without a developer API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights" ], "source_metadata": { "name": "Solar Open 2", "description": "Open-weight 250B-A15B large language model built for agentic use cases such as office productivity, document-intensive work, and coding", "family": "solar-open", "release_date": "2026-07-22", "last_updated": null, "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": null, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": null }, "reported_open_weights": true, "reported_license": "other", "links": [], "weights": [] }, "licenses": [ "LicenseRef-Upstage-Solar" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The Upstage Solar License adopts the grants of the Apache License 2.0, including commercial use, and adds Section 4(e): anyone who distributes or makes available a model created, trained, fine-tuned, or otherwise improved using the weights must begin its name with \"Solar\", prominently display \"Built with Solar\", and provide the license alongside it. Upstage's announcement confirms that commercial use and derivative models through fine-tuning and distillation are permitted. A naming and branding condition on every derivative model departs from an open-source posture, so the release is source-available rather than open source.", "license_evidence": [ { "kind": "web_terms", "license_id": "LicenseRef-Upstage-Solar", "scope": "upstage/Solar-Open2-250B weights, configuration, tokenizer, and documentation, and any derivative AI model created, trained, fine-tuned, or otherwise improved using them. The NotaAI quantizations are published in separate repositories.", "url": "https://huggingface.co/upstage/Solar-Open2-250B/blob/main/LICENSE", "verified_at": "2026-09-27" } ], "access_boundary": "This record represents the upstage/Solar-Open2-250B release announced July 22, 2026. The predecessor Solar Open 100B, the NotaAI INT4, INT4-GlobalPruned, and NVFP4 quantizations, Upstage's forks of Transformers and vLLM, the proprietary Solar Pro and Solar Mini API models, and the Upstage Playground trial that ended July 31, 2026 remain separate boundaries.", "strengths": [ "Developer-published BF16 safetensors checkpoint on the Upstage Hugging Face organization, with the technical report and license file in the same repository and official NotaAI INT4 and NVFP4 quantizations linked from the model card", "First-party model card documents the architecture, parameter counts, context length, training tokens and hardware, reasoning modes, tool-call parsing, and Transformers and vLLM serving paths, including connecting Claude Code and Hermes Agent to a local server" ], "tradeoffs": [ "The Upstage Solar License requires every distributed derivative model to begin its name with \"Solar\" and to display \"Built with Solar\"", "Serving depends on Upstage's forks of Transformers and vLLM on NVIDIA GPUs, and the documented examples assume eight GPUs with at least 141 GB each; no developer API or third-party hosted path is documented", "No versioning or deprecation posture is stated for the Solar Open line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 6, "serving_reach": 4, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.36 }, "evidence": [ { "kind": "web", "label": "Official Upstage Solar Open 2 announcement: release, architecture, license, and channels", "url": "https://www.upstage.ai/blog/en/solar-open-2", "verified_at": "2026-09-27" }, { "kind": "web", "label": "Developer-published upstage/Solar-Open2-250B weights repository and model card", "url": "https://huggingface.co/upstage/Solar-Open2-250B", "verified_at": "2026-09-27" }, { "kind": "web", "label": "Upstage Solar License file with the Section 4(e) derivative-model naming conditions", "url": "https://huggingface.co/upstage/Solar-Open2-250B/blob/main/LICENSE", "verified_at": "2026-09-27" }, { "kind": "web", "label": "Model configuration giving the 1,048,576-token maximum position embeddings", "url": "https://huggingface.co/upstage/Solar-Open2-250B/raw/main/config.json", "verified_at": "2026-09-27" }, { "kind": "web", "label": "Upstage Console model catalog, which lists no Solar Open 2 API model", "url": "https://console.upstage.ai/docs/models", "verified_at": "2026-09-27" } ], "metadata_verified_at": "2026-09-27", "verified_at": "2026-09-27" }, { "id": "model-upstage-solar-pro2", "source_id": "upstage/solar-pro2", "name": "Solar Pro 2", "developer": "Upstage", "url": "https://console.upstage.ai/docs/models/solar-pro-2", "description": "Upstage's 31-billion-parameter Solar Pro 2 release, served as dated snapshot API models with a tokenizer-only public artifact.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Solar Pro 2", "description": "Flagship model for demanding analysis, coding, and production agent workflows", "family": "solar-pro", "release_date": "2025-05-20", "last_updated": "2025-05-20", "knowledge_cutoff": "2025-03", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 65536, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model-weight license exists: no weights are published, only a tokenizer, and use is governed by the Upstage Terms of Use, which grant a non-transferable limited service license and reserve all intellectual property. The release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of solar-pro2 snapshots with no published model license", "kind": "web_terms", "url": "https://www.upstage.ai/terms-of-service", "verified_at": "2026-09-25" } ], "access_boundary": "This record represents the July 2025 Solar Pro 2 line served as dated snapshots (250710, 250909, 251215) with a tokenizer-only public artifact. The Solar Pro 3 successor, other Upstage lines, and the Upstage platform itself are separate boundaries.", "strengths": [ "Console docs pin three dated snapshots with changelogs for the exact model IDs", "First-party Console serving with AWS Marketplace and on-premise paths" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 2, "deployment_portability": 6, "serving_reach": 6, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 5.58 }, "evidence": [ { "kind": "web", "label": "Official Solar Pro 2 docs with dated snapshots", "url": "https://console.upstage.ai/docs/models/solar-pro-2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Solar Pro 2 launch announcement (July 2025)", "url": "https://www.upstage.ai/news/solar-pro-2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/upstage/solar-pro2.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-upstage-solar-pro3", "source_id": "upstage/solar-pro3", "name": "Solar Pro 3", "developer": "Upstage", "url": "https://console.upstage.ai/docs/models/solar-pro-3", "description": "Upstage's 102-billion-parameter Mixture-of-Experts Solar Pro 3 release with 12B active parameters, served as dated snapshot API models with a tokenizer-only public artifact.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Solar Pro 3", "description": "Flagship model for demanding analysis, coding, and production agent workflows", "family": "solar-pro", "release_date": "2026-01", "last_updated": "2026-01", "knowledge_cutoff": "2025-03", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No model-weight license exists: no weights are published, only a tokenizer whose own Solar license is artifact-scoped and does not grant model rights. Use is governed by the Upstage Terms of Use, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of solar-pro3 snapshots with no published model license", "kind": "web_terms", "url": "https://www.upstage.ai/terms-of-service", "verified_at": "2026-09-25" } ], "access_boundary": "This record represents the January 2026 Solar Pro 3 line served as dated snapshots (260126, 260323) with a tokenizer-only public artifact. The Solar Pro 2 predecessor, the Pro 4 successor, other Upstage lines, and the platform itself are separate boundaries.", "strengths": [ "Console docs pin two dated snapshots with a deprecation pointer for the exact model IDs", "First-party Console serving with dated launch announcements" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 2, "deployment_portability": 4, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 5.06 }, "evidence": [ { "kind": "web", "label": "Official Solar Pro 3 docs with dated snapshots", "url": "https://console.upstage.ai/docs/models/solar-pro-3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Solar Pro 3 launch announcement (January 2026)", "url": "https://www.upstage.ai/blog/en/solar-pro-3-0127", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/upstage/solar-pro3.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-upstage-solar-pro4", "source_id": "upstage/solar-pro4", "name": "Solar Pro 4", "developer": "Upstage", "url": "https://console.upstage.ai/docs/models/solar-pro-4", "description": "Upstage's August 2026 Solar Pro 4 agentic release with 512K context, served as a managed API with no public artifact of any kind.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Solar Pro 4", "description": "Upstage's flagship model, specialized for agentic use", "family": "solar-pro", "release_date": "2026-08-06", "last_updated": "2026-08-06", "knowledge_cutoff": "2026-02", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 524288, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No license of any kind is published for this closed commercial line: no weights, no tokenizer, no model terms beyond the Upstage Terms of Use. The release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of solar-pro4 with no published license", "kind": "web_terms", "url": "https://www.upstage.ai/terms-of-service", "verified_at": "2026-09-25" } ], "access_boundary": "This record represents the August 2026 Solar Pro 4 line served as the solar-pro4 API model with 512K context and 128K max output. Earlier Solar generations, other Upstage lines, and the platform itself are separate boundaries.", "strengths": [ "Dated August 2026 launch pins the line with full specs and Console serving", "512K context with reasoning traces plus Studio, SolarChat, and on-premise surfaces" ], "tradeoffs": [ "No downloadable artifact of any kind exists first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 1, "deployment_portability": 5, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 5.1 }, "evidence": [ { "kind": "web", "label": "Official Solar Pro 4 docs with specs and serving", "url": "https://console.upstage.ai/docs/models/solar-pro-4", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Solar Pro 4 launch announcement (August 2026)", "url": "https://www.upstage.ai/blog/en/solar-pro-4", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/upstage/solar-pro4.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-vispark-vision-large", "source_id": "vispark/vision-large", "name": "Vision Large", "developer": "Vispark", "url": "https://vispark.in/ai", "description": "Vispark's Vision Large, the most capable tier of the Vision foundational model line, a multimodal model with text output and 1M context served through the Vispark Lab API.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Vision Large", "description": "Most capable Vision model for complex reasoning, detailed media analysis, and structured output over a 1M-token context window.", "family": null, "release_date": "2024-05-15", "last_updated": "2026-09", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Vispark-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Vision Large is served, not distributed, under the Vispark service terms; no weights are released.", "license_evidence": [ { "license_id": "LicenseRef-Vispark-Service-Terms", "scope": "Vision Large served through the Vispark Lab API under Vispark service terms", "kind": "web_terms", "url": "https://vispark.in/terms", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Vision Large tier of the Vispark Vision line. The Vision Medium and Vision Small tiers are separate records, and the consumer Chat product with image and voice generation is a separate service boundary.", "strengths": [ "First-party model page, lab, and dated launch blog establish the line identity", "Live API with a documented 1M-token context and three-tier pricing" ], "tradeoffs": [ "No fixed snapshot identity beyond the dateless vision-large tier name", "API output is text while the consumer product advertises image and voice generation, which needs careful scoping" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/vispark/vision-large.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://vispark.in/ai", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-vispark-vision-medium", "source_id": "vispark/vision-medium", "name": "Vision Medium", "developer": "Vispark", "url": "https://vispark.in/ai", "description": "Vispark's Vision Medium, the balanced tier of the Vision foundational model line, a multimodal model with text output served through the Vispark Lab API.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Vision Medium", "description": "Balanced multimodal model pairing a 1M-token context window with deeper reasoning for analysis, content creation, and tool use across text, image, audio, video, and PDF inputs.", "family": null, "release_date": "2024-05-15", "last_updated": "2026-09", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Vispark-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Vision Medium is served, not distributed, under the Vispark service terms; no weights are released.", "license_evidence": [ { "license_id": "LicenseRef-Vispark-Service-Terms", "scope": "Vision Medium served through the Vispark Lab API under Vispark service terms", "kind": "web_terms", "url": "https://vispark.in/terms", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Vision Medium tier of the Vispark Vision line. The Vision Large and Vision Small tiers are separate records, and the consumer Chat product is a separate service boundary.", "strengths": [ "First-party model page and dated launch blog establish the line identity", "Live API with documented pricing for a distinct balanced tier" ], "tradeoffs": [ "No fixed snapshot identity beyond the dateless vision-medium tier name", "API output is text while the consumer product advertises image and voice generation" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/vispark/vision-medium.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://vispark.in/ai", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-vispark-vision-small", "source_id": "vispark/vision-small", "name": "Vision Small", "developer": "Vispark", "url": "https://vispark.in/ai", "description": "Vispark's Vision Small, the fastest and lowest-cost tier of the Vision foundational model line, a multimodal model with text output served through the Vispark Lab API.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Vision Small", "description": "Fast, low-cost multimodal model for understanding text, images, audio, video, and PDFs, with tool calling and a 1M-token context window.", "family": null, "release_date": "2024-05-15", "last_updated": "2026-09", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 65536 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Vispark-Service-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Vision Small is served, not distributed, under the Vispark service terms; no weights are released.", "license_evidence": [ { "license_id": "LicenseRef-Vispark-Service-Terms", "scope": "Vision Small served through the Vispark Lab API under Vispark service terms", "kind": "web_terms", "url": "https://vispark.in/terms", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Vision Small tier of the Vispark Vision line. The Vision Large and Vision Medium tiers are separate records, and the consumer Chat product is a separate service boundary.", "strengths": [ "First-party model page and dated launch blog establish the line identity", "Live API with a documented low-cost tier for fast responses" ], "tradeoffs": [ "No fixed snapshot identity beyond the dateless vision-small tier name", "API output is text while the consumer product advertises image and voice generation" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.74 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/vispark/vision-small.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://vispark.in/ai", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-vivgrid-viv-fast", "source_id": "vivgrid/viv-fast", "name": "Viv Fast", "developer": "Vivgrid", "url": "https://vivgrid.com/docs/models/viv-fast", "description": "Vivgrid's Viv Fast, a closed-weight multimodal model with 1M context and 256K output marketed for low-cost tool-calling, served through the Vivgrid API with a dated September 21, 2026 docs identity.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Viv Fast", "description": "Fast coding model", "family": null, "release_date": "2026-09-21", "last_updated": "2026-09-21", "knowledge_cutoff": "2026-09", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": false }, "limits": { "context": 1000000, "input": null, "output": 256000 }, "reported_open_weights": false, "reported_license": null, "links": [ { "label": "Model docs", "url": "https://docs.vivgrid.com/models/viv-fast" } ], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Viv Fast is served, not distributed, under the Vivgrid service terms; no weights or license file exist and the provenance of the underlying weights is not first-party disclosed.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "Viv Fast served through the Vivgrid API under Vivgrid service terms", "kind": "web_terms", "url": "https://vivgrid.com/docs/models/viv-fast", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the Viv Fast API model documented on Vivgrid's model page. Vivgrid is primarily a gateway provider, so the boundary keeps the reviewed API identity distinct from any underlying third-party model.", "strengths": [ "Dated first-party docs page with concrete specs, pricing, and context limits", "A low-cost API tier with a documented 1M-context serving surface" ], "tradeoffs": [ "Identity rests on a docs page alone with no announcement or model card", "Provenance is unverified because Vivgrid is a gateway and the underlying model is not disclosed" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 6, "overall": 3.6 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/vivgrid/viv-fast.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://vivgrid.com/docs/models/viv-fast", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-writer-palmyra-x4", "source_id": "writer/palmyra-x4", "name": "Palmyra X4", "developer": "Writer", "url": "https://dev.writer.com/home/models", "description": "Writer's Palmyra X4, a 128K-context tool-calling and RAG-focused language model launched September 2024, served through the Writer API and Amazon Bedrock and scheduled for deprecation on December 14, 2026.", "model_type": "language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Palmyra X4", "description": "Enterprise language model for workflow automation, coding, data analysis, and tool use", "family": "palmyra", "release_date": "2024-10-09", "last_updated": "2025-04-28", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 4096 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Writer-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Palmyra X4 is served, not distributed; the governing terms are Writer's API and product terms, with a model card on Amazon Bedrock documenting the serving lifecycle and a published deprecation date of December 14, 2026.", "license_evidence": [ { "license_id": "LicenseRef-Writer-API-Terms", "scope": "Palmyra X4 served API model under Writer API and product terms; Bedrock card documents the serving lifecycle", "kind": "web_terms", "url": "https://writer.com/legal/", "verified_at": "2026-10-03" } ], "access_boundary": "This record represents the Palmyra X4 API model line launched September 2024 with the palmyra-x4 API identifier, served through Writer and Amazon Bedrock, and scheduled for deprecation on December 14, 2026. Writer documents Palmyra X6 as the migration target. Palmyra X5 is a separate line also scheduled for deprecation.", "strengths": [ "Dated first-party identity with a published deprecation date and an explicit migration path to Palmyra X6", "Available on two channels (Writer API and Amazon Bedrock) with documented tool-calling and structured-output support" ], "tradeoffs": [ "Closed weights with proprietary terms only", "The line is scheduled for deprecation with continued support only through Writer's published policy window" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/writer/palmyra-x4.toml", "verified_at": "2026-10-03" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://dev.writer.com/home/models", "verified_at": "2026-10-03" } ], "metadata_verified_at": "2026-10-03", "verified_at": "2026-10-03" }, { "id": "model-writer-palmyra-x5", "source_id": "writer/palmyra-x5", "name": "Palmyra X5", "developer": "Writer", "url": "https://dev.writer.com/home/models", "description": "Writer's Palmyra X5, launched April 28, 2025, a 1M-context multimodal model taking text and image input and returning text or structured output, served through the Writer API and Amazon Bedrock and scheduled for deprecation on December 14, 2026.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Palmyra X5", "description": "Reasoning model for deliberate analysis, multi-step problem solving, and tool use", "family": "palmyra", "release_date": "2025-04-28", "last_updated": "2025-04-28", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Writer-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Palmyra X5 is served, not distributed; the governing terms are Writer's API and product terms, with dated launch and serving documentation, a Bedrock model card, and a published deprecation date of December 14, 2026.", "license_evidence": [ { "license_id": "LicenseRef-Writer-API-Terms", "scope": "Palmyra X5 served API model under Writer API and product terms; Bedrock card documents the serving lifecycle", "kind": "web_terms", "url": "https://writer.com/legal/", "verified_at": "2026-10-03" } ], "access_boundary": "This record represents the Palmyra X5 API model line launched April 28, 2025 with the palmyra-x5 API identifier, served through Writer and Amazon Bedrock and scheduled for deprecation on December 14, 2026. Writer documents Palmyra X6 as the migration target. Palmyra X4 is a separate deprecating line.", "strengths": [ "Dated first-party launch identity with a published deprecation date and migration path to Palmyra X6", "1M-token context with image input and sub-second tool-calling for agentic workloads on two channels" ], "tradeoffs": [ "Closed weights with proprietary terms only", "The line is scheduled for deprecation even though it remains the documented vision model for some API examples" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 4.28 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/writer/palmyra-x5.toml", "verified_at": "2026-10-03" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://dev.writer.com/home/models", "verified_at": "2026-10-03" } ], "metadata_verified_at": "2026-10-03", "verified_at": "2026-10-03" }, { "id": "model-writer-palmyra-x6", "source_id": null, "name": "Palmyra X6", "developer": "Writer", "url": "https://dev.writer.com/home/models", "description": "Writer's Palmyra X6, launched August 13, 2026, a 1M-context agentic language model for long-horizon enterprise workflows with tool use and sub-agents, served as the default model on the Writer API.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Palmyra X6", "description": "Agentic model for WRITER Agent and long-horizon enterprise workflows with sub-agents, grounding, and MCP tool use", "family": "palmyra", "release_date": "2026-08-13", "last_updated": "2026-10-03", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 8192 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Writer-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Palmyra X6 is served, not distributed; the governing terms are Writer's API and product terms, with dated launch documentation and a published model catalog covering pricing, context limits, and family-level deprecation policy.", "license_evidence": [ { "license_id": "LicenseRef-Writer-API-Terms", "scope": "Palmyra X6 served API model under Writer API and product terms", "kind": "web_terms", "url": "https://writer.com/legal/", "verified_at": "2026-10-03" } ], "access_boundary": "This record represents the Palmyra X6 API model line launched August 13, 2026 with the palmyra-x6 API identifier and documented as Writer's default agentic model on the Writer API. Palmyra X4 and X5 are separate deprecating lines.", "strengths": [ "Dated first-party launch identity with a published model catalog, pricing, and family-level deprecation policy", "Documented agentic capabilities with 1M-token context, tool use, and sub-agents on the Writer API" ], "tradeoffs": [ "Closed weights with proprietary terms only", "Writer documents first-party API serving for this line; no Amazon Bedrock model ID is listed alongside X4 and X5" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 8, "documentation_provenance": 8, "overall": 4.24 }, "evidence": [ { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://dev.writer.com/home/models", "verified_at": "2026-10-03" }, { "kind": "web", "label": "Palmyra X6 launch announcement", "url": "https://writer.com/engineering/palmyra-x6/", "verified_at": "2026-10-03" } ], "metadata_verified_at": "2026-10-03", "verified_at": "2026-10-03" }, { "id": "model-xai-grok-4-20-0309-non-reasoning", "source_id": "xai/grok-4.20-0309-non-reasoning", "name": "Grok 4.20 (Non-Reasoning)", "developer": "xAI", "url": "https://docs.x.ai/developers/models/grok-4.20-0309-non-reasoning", "description": "An API-served image-and-text Grok snapshot release with the fixed dated identity grok-4.20-0309-non-reasoning and a 1,000,000-token context, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Grok 4.20 (Non-Reasoning)", "description": "Grok model for agentic tool use, reasoning, coding, and live assistance", "family": "grok", "release_date": "2026-03-09", "last_updated": "2026-03-09", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 30000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "xAI provides the Grok release as a managed API service under its Terms of Service - Enterprise, which incorporate the Acceptable Use Policy by reference. The terms permit customers to build and distribute integrations on the API, while the policy prohibits reverse engineering or distilling the models and using the service or its output to develop competing models; no model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-18", "scope": "Business access to the Grok release through the xAI API and related xAI services, including the right to build and distribute Bundled Services" }, { "license_id": "LicenseRef-xAI-AUP", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-18", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into the xAI Terms of Service - Enterprise" } ], "access_boundary": "This record represents the Grok 4.20 non-reasoning snapshot release of March 9, 2026 under the fixed API model name grok-4.20-0309-non-reasoning. The reasoning variant, the multi-agent sibling, Oracle OCI and Cloudflare hosted paths, consumer apps, and later Grok releases remain separate service or product boundaries.", "strengths": [ "xAI publishes a release-specific model page with a fixed dated model name, documented modalities, context, pricing, and regions", "Oracle OCI and Cloudflare publish host documentation naming the exact model ID" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The line is already superseded by later Grok releases with no retirement notice for this snapshot" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 6, "overall": 3.86 }, "evidence": [ { "kind": "web", "label": "Official Grok 4.20 0309 non-reasoning model documentation page", "url": "https://docs.x.ai/developers/models/grok-4.20-0309-non-reasoning", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Oracle OCI Generative AI release notes naming the exact model", "url": "https://docs.oracle.com/en-us/iaas/releasenotes/generative-ai/xAI-grok-4-20.htm", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Cloudflare AI model page for the exact model", "url": "https://developers.cloudflare.com/ai/models/xai/grok-4.20-0309-non-reasoning", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xai/grok-4.20-0309-non-reasoning.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xai-grok-4-20-0309-reasoning", "source_id": "xai/grok-4.20-0309-reasoning", "name": "Grok 4.20 (Reasoning)", "developer": "xAI", "url": "https://docs.x.ai/developers/models/grok-4.20-0309-reasoning", "description": "An API-served image-and-text reasoning Grok snapshot release with the fixed dated identity grok-4.20-0309-reasoning and a 1,000,000-token context, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Grok 4.20 (Reasoning)", "description": "Reasoning Grok for document-heavy analysis and long-horizon tool use", "family": "grok", "release_date": "2026-03-09", "last_updated": "2026-03-09", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 30000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "xAI provides the Grok release as a managed API service under its Terms of Service - Enterprise, which incorporate the Acceptable Use Policy by reference. The terms permit customers to build and distribute integrations on the API, while the policy prohibits reverse engineering or distilling the models and using the service or its output to develop competing models; no model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-18", "scope": "Business access to the Grok release through the xAI API and related xAI services, including the right to build and distribute Bundled Services" }, { "license_id": "LicenseRef-xAI-AUP", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-18", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into the xAI Terms of Service - Enterprise" } ], "access_boundary": "This record represents the Grok 4.20 reasoning snapshot release of March 9, 2026 under the fixed API model name grok-4.20-0309-reasoning. The non-reasoning variant, the multi-agent sibling, Oracle OCI hosted paths, consumer apps, and later Grok releases remain separate service or product boundaries.", "strengths": [ "xAI publishes a release-specific model page with a fixed dated model name, documented modalities, context, pricing, and regions", "xAI publishes a system card and Oracle OCI publishes host documentation naming the exact model ID" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The line is already superseded by later Grok releases with no retirement notice for this snapshot" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 6, "overall": 3.86 }, "evidence": [ { "kind": "web", "label": "Official Grok 4.20 0309 reasoning model documentation page", "url": "https://docs.x.ai/developers/models/grok-4.20-0309-reasoning", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Grok 4.20 system card", "url": "https://data.x.ai/2026-04-07-grok-4-20-model-card.pdf", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Oracle OCI Generative AI release notes naming the exact model", "url": "https://docs.oracle.com/en-us/iaas/releasenotes/generative-ai/xAI-grok-4-20.htm", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xai/grok-4.20-0309-reasoning.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xai-grok-4-3", "source_id": "xai/grok-4.3", "name": "Grok 4.3", "developer": "xAI", "url": "https://docs.x.ai/developers/models/grok-4.3", "description": "An API-served image-and-text Grok release launched in April 2026 behind the rotating grok-4.3 and grok-4.3-latest aliases, with a 1,000,000-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "Grok 4.3", "description": "xAI's default Grok for chat, coding, agentic tools, and lower hallucination risk", "family": "grok", "release_date": "2026-04-17", "last_updated": "2026-04-17", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 30000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "xAI provides the Grok release as a managed API service under its Terms of Service - Enterprise, which incorporate the Acceptable Use Policy by reference. The terms permit customers to build and distribute integrations on the API, while the policy prohibits reverse engineering or distilling the models and using the service or its output to develop competing models; no model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-18", "scope": "Business access to the Grok release through the xAI API and related xAI services, including the right to build and distribute Bundled Services" }, { "license_id": "LicenseRef-xAI-AUP", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-18", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into the xAI Terms of Service - Enterprise" } ], "access_boundary": "This record represents the Grok 4.3 release launched April 17, 2026 and served under the rotating API model names grok-4.3 and grok-4.3-latest. The 4.2 and 4.20 generations, consumer apps, Grok Build, gateway partners, and later Grok releases remain separate service or product boundaries.", "strengths": [ "xAI publishes a release-specific model page with documented modalities, context, pricing, and three first-party regions" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "xAI publishes no dated snapshot, launch announcement, or release-notes entry for this release" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 5, "documentation_provenance": 5, "overall": 3.46 }, "evidence": [ { "kind": "web", "label": "Official Grok 4.3 model documentation page", "url": "https://docs.x.ai/developers/models/grok-4.3", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xai/grok-4.3.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xai-grok-4-5", "source_id": "xai/grok-4.5", "name": "Grok 4.5", "developer": "xAI", "url": "https://docs.x.ai/developers/models/grok-4.5", "description": "An API-served image-and-text Grok release launched July 2026 for coding, agentic tasks, and knowledge work, with a 500,000-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Grok 4.5", "description": "xAI's Grok model for chat, coding, agentic tools, and lower hallucination risk", "family": "grok", "release_date": "2026-07-08", "last_updated": "2026-07-08", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 500000, "input": null, "output": 500000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "xAI provides the Grok release as a managed API service under its Terms of Service - Enterprise, which incorporate the Acceptable Use Policy by reference. The terms permit customers to build and distribute integrations on the API, while the policy prohibits reverse engineering or distilling the models and using the service or its output to develop competing models; no model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-18", "scope": "Business access to the Grok release through the xAI API and related xAI services, including the right to build and distribute Bundled Services" }, { "license_id": "LicenseRef-xAI-AUP", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-18", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into the xAI Terms of Service - Enterprise" } ], "access_boundary": "This record represents the Grok 4.5 release announced July 2026 and served under the rotating API model names grok-4.5 and grok-4.5-latest. The Grok Build and Cursor integrations named in the announcement, consumer apps, and later Grok releases remain separate service or product boundaries.", "strengths": [ "xAI publishes a release-specific model page and a dated launch announcement naming the Grok Build and Cursor integrations", "Documented modalities, context, pricing, regions, and EU availability rollout" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "xAI publishes no dated snapshot ID, so the fixed identity rests on the launch documentation alone" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 4.0 }, "evidence": [ { "kind": "web", "label": "Official Grok 4.5 model documentation page", "url": "https://docs.x.ai/developers/models/grok-4.5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official xAI Grok 4.5 release announcement", "url": "https://x.ai/news/grok-4-5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xai/grok-4.5.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xai-grok-4-6", "source_id": "xai/grok-4.6", "name": "Grok 4.6", "developer": "xAI", "url": "https://docs.x.ai/developers/models/grok-4.6", "description": "An API-served image-and-text Grok release documented on xAI's developer model page and offered as a partner model on Google Cloud Vertex AI, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Grok 4.6", "description": "xAI's frontier model for long-running agents, coding, knowledge work, and visual projects", "family": "grok", "release_date": "2026-08-12", "last_updated": "2026-08-12", "knowledge_cutoff": "2026-02-01", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 500000, "input": null, "output": 500000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "xAI provides Grok 4.6 as a managed API service under its Terms of Service - Enterprise, which incorporate the Acceptable Use Policy by reference. The terms permit customers to build and distribute integrations on the API, while the policy prohibits reverse engineering or distilling the models and using the service or its output to develop competing models; no model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-11", "scope": "Business access to Grok 4.6 through the xAI API and related xAI services, including the right to build and distribute Bundled Services" }, { "license_id": "LicenseRef-xAI-AUP", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-11", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into the xAI Terms of Service - Enterprise" } ], "access_boundary": "This record represents the Grok 4.6 release served under the xAI API model name grok-4.6, with no dated snapshot ID published. xAI's own API, Google Cloud Vertex AI, Microsoft Foundry, the fast variant, Grok consumer apps, Grok Build, gateway partners, and later Grok releases remain separate service or product boundaries.", "strengths": [ "xAI publishes a release-specific model page with a fixed model name, documented modalities, release date, knowledge cutoff, and three first-party regions", "xAI's own docs document an independent Google Cloud Vertex AI partner-model path using the ID xai/grok-4.6" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "xAI publishes no dated snapshot ID, per-model lifecycle status, or general retirement-notice policy, and prior retirements silently rerouted requests to replacement models" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 6, "overall": 3.74 }, "evidence": [ { "kind": "web", "label": "Official Grok 4.6 model documentation page", "url": "https://docs.x.ai/developers/models/grok-4.6", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official xAI Grok 4.6 release announcement", "url": "https://x.ai/news/grok-4-6", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/xai/grok-4.6.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-xai-grok-build-0-1", "source_id": "xai/grok-build-0.1", "name": "Grok Build 0.1", "developer": "xAI", "url": "https://docs.x.ai/developers/models/grok-build-0.1", "description": "An API-served coding model in public beta, the same model that powers Grok Build, with a 256,000-token context and no downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api", "third_party_hosting" ], "source_metadata": { "name": "Grok Build 0.1", "description": "Fast Grok coding model tuned for agentic engineering and iterative edits", "family": "grok-build", "release_date": "2026-04-16", "last_updated": "2026-04-16", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 256000, "input": null, "output": 256000 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-xAI-Enterprise-Terms", "LicenseRef-xAI-AUP" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "xAI provides the Grok release as a managed API service under its Terms of Service - Enterprise, which incorporate the Acceptable Use Policy by reference. The terms permit customers to build and distribute integrations on the API, while the policy prohibits reverse engineering or distilling the models and using the service or its output to develop competing models; no model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-xAI-Enterprise-Terms", "kind": "web_terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "verified_at": "2026-09-18", "scope": "Business access to the Grok release through the xAI API and related xAI services, including the right to build and distribute Bundled Services" }, { "license_id": "LicenseRef-xAI-AUP", "kind": "web_terms", "url": "https://x.ai/legal/acceptable-use-policy", "verified_at": "2026-09-18", "scope": "Acceptable-use, anti-distillation, and competing-model restrictions incorporated into the xAI Terms of Service - Enterprise" } ], "access_boundary": "This record represents the Grok Build 0.1 model released in public beta and served under the API model name grok-build-0.1. The xai-org/grok-build harness repository, Grok Build the product, the grok-code-fast aliases, gateway partners, and later Grok releases remain separate service or product boundaries.", "strengths": [ "xAI publishes a model page and a dated announcement, and Vercel AI Gateway publishes host documentation naming the exact model", "The announcement names the OpenRouter and Vercel AI Gateway availability paths" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "The line carries early-access status and its aliases already point to later Grok releases" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.88 }, "evidence": [ { "kind": "web", "label": "Official Grok Build 0.1 model documentation page", "url": "https://docs.x.ai/developers/models/grok-build-0.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official xAI Grok Build 0.1 release announcement", "url": "https://x.ai/news/grok-build-0-1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Vercel AI Gateway model page for the exact model", "url": "https://vercel.com/ai-gateway/models/grok-build-0.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xai/grok-build-0.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xiaomi-mimo-v2-flash", "source_id": "xiaomi/mimo-v2-flash", "name": "MiMo-V2-Flash", "developer": "Xiaomi", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash", "description": "Xiaomi's 309-billion-parameter Mixture-of-Experts release with 15B active parameters for efficient inference, published as MIT-claimed open weights with a launch-era managed API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiMo-V2-Flash", "description": "MiMo flash model for fast multimodal assistance and agent workflows", "family": "mimo", "release_date": "2025-12-16", "last_updated": "2026-02-04", "knowledge_cutoff": "2024-12-01", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 65536 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The weight repository carries no LICENSE file; the MIT grant rests on concordant first-party declarations that the weights and inference code are fully open-sourced under the MIT license, with no use restrictions attached. The missing file caps clarity below a pinned-license grant.", "license_evidence": [ { "license_id": "MIT", "scope": "MiMo-V2-Flash weights and inference code as declared in the official release announcement", "kind": "web_terms", "url": "https://mimo.mi.com/docs/en-US/news/previous-news/news20251216", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the December 2025 MiMo-V2-Flash release at Hugging Face commit 1afd314a2406c282e0956375c34a676501c78649 (last modified 2026-07-09), with three dated 2026 update notes. The Flash-Base sibling, the V2 Pro and Omni lines, later MiMo generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 309B MoE checkpoint with Base sibling and MTP weights, all MIT-claimed", "Release announcement pins the December 2025 launch with API pricing and weight links" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on announcement prose and tags", "The launch-era free API period has long lapsed" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 7, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 7.28 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official MiMo-V2-Flash release announcement (December 2025)", "url": "https://mimo.mi.com/docs/en-US/news/previous-news/news20251216", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xiaomi/mimo-v2-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xiaomi-mimo-v2-omni", "source_id": "xiaomi/mimo-v2-omni", "name": "MiMo-V2-Omni", "developer": "Xiaomi", "url": "https://mimo.xiaomi.com/mimo-v2-omni", "description": "Xiaomi's all-modal V2 release fusing text, vision, and speech encoders, served as the mimo-v2-omni API model with no public weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "MiMo-V2-Omni", "description": "MiMo omni model for text, image, video, audio, and agents", "family": "mimo", "release_date": "2026-03-18", "last_updated": "2026-03-18", "knowledge_cutoff": "2024-12", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 262144, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No license of any kind is published for this API-only line: no weight repository exists under any Xiaomi ID, and neither launch nor docs page states license terms. Use is governed by Xiaomi platform API terms, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of mimo-v2-omni with no published license", "kind": "web_terms", "url": "https://mimo.xiaomi.com/mimo-v2-omni", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the March 2026 mimo-v2-omni line with native text, image, video, and audio input and text output, served through Xiaomi's API platform. The Pro and Flash V2 siblings, later omni-modal generations, and the Xiaomi platform itself are separate boundaries.", "strengths": [ "Dedicated first-party launch and docs pages detail per-modality input including 10-hour audio", "Managed API serving with published input and output pricing" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 3.78 }, "evidence": [ { "kind": "web", "label": "Official mimo-v2-omni launch page (March 2026)", "url": "https://mimo.xiaomi.com/mimo-v2-omni", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official mimo-v2-omni release announcement", "url": "https://mimo.mi.com/docs/en-US/news/previous-news/v2-omni-release", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xiaomi/mimo-v2-omni.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xiaomi-mimo-v2-pro", "source_id": "xiaomi/mimo-v2-pro", "name": "MiMo-V2-Pro", "developer": "Xiaomi", "url": "https://mimo.xiaomi.com/mimo-v2-pro", "description": "Xiaomi's trillion-parameter flagship for agentic workloads with 42B active parameters and 1M context, served as the mimo-v2-pro API model with no public weights.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "MiMo-V2-Pro", "description": "Earlier MiMo Pro model for multimodal agents, reasoning, and code tasks", "family": "mimo", "release_date": "2026-03-18", "last_updated": "2026-03-18", "knowledge_cutoff": "2024-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-Proprietary" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "No license of any kind is published for this API-only line: no weight repository exists under any Xiaomi ID, and neither launch nor docs page states license terms. Use is governed by Xiaomi platform API terms, so the release is proprietary.", "license_evidence": [ { "license_id": "LicenseRef-Proprietary", "scope": "API-only commercial distribution of mimo-v2-pro with no published license", "kind": "web_terms", "url": "https://mimo.xiaomi.com/mimo-v2-pro", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the March 2026 mimo-v2-pro flagship line (1T total, 42B active, 1M context) served through Xiaomi's API platform. The Flash and Omni V2 siblings, later MiMo generations, and the Xiaomi platform itself are separate boundaries.", "strengths": [ "Dedicated first-party launch and docs pages pin the March 2026 release with API pricing", "Trillion-parameter scale with 1M context documented for agentic workflows" ], "tradeoffs": [ "No downloadable weights exist anywhere first-party for this line", "No scrutable license grant of any kind; platform terms govern entirely" ], "score_profile": "model_access", "score": { "license_clarity": 3, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 5, "lifecycle_transparency": 6, "documentation_provenance": 7, "overall": 3.64 }, "evidence": [ { "kind": "web", "label": "Official mimo-v2-pro launch page (March 2026)", "url": "https://mimo.xiaomi.com/mimo-v2-pro", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official mimo-v2-pro release announcement with API pricing", "url": "https://mimo.mi.com/docs/en-US/news/previous-news/v2-pro-release", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xiaomi/mimo-v2-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xiaomi-mimo-v2-5", "source_id": "xiaomi/mimo-v2.5", "name": "MiMo-V2.5", "developer": "Xiaomi", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5", "description": "Xiaomi's 310-billion-parameter native omni-modal release with 15B active parameters, published as MIT-claimed open weights with managed API, Studio, and Token Plan serving.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiMo-V2.5", "description": "Open MiMo model for multimodal coding agents and long-context automation", "family": "mimo", "release_date": "2026-04-22", "last_updated": "2026-04-22", "knowledge_cutoff": "2024-12", "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The weight repositories carry no LICENSE file; the MIT grant rests on the official open-sourcing announcement, which states both models including Base are fully open-sourced under the permissive MIT license with free commercial use and no additional authorization. The missing file caps clarity below a pinned-license grant.", "license_evidence": [ { "license_id": "MIT", "scope": "MiMo-V2.5 and Base weights as declared in the official open-sourcing announcement", "kind": "web_terms", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the April 2026 MiMo-V2.5 release at Hugging Face commit 63651580ca774f8504f676040460aed3e1244ac1 (last modified 2026-07-09), with native text, image, video, and audio input and 1M context, together with its API Platform, AI Studio, and Token Plan serving. The Base sibling, the V2.5-Pro line, later MiMo generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated 310B omni-modal checkpoint with Base sibling, all MIT-claimed with commercial use allowed", "First-party API Platform, AI Studio, and Token Plan serving for the exact model ID" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on announcement prose and tags", "310B total parameters need multi-GPU serving for self-operation" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.18 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official MiMo-V2.5 open-sourcing announcement with the MIT grant", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official MiMo-V2.5 launch page (April 2026)", "url": "https://mimo.xiaomi.com/mimo-v2-5/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xiaomi/mimo-v2.5.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xiaomi-mimo-v2-5-pro", "source_id": "xiaomi/mimo-v2.5-pro", "name": "MiMo-V2.5-Pro", "developer": "Xiaomi", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro", "description": "Xiaomi's 1.02-trillion-parameter text release with 42B active parameters, published as MIT-claimed open weights with managed API serving.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiMo-V2.5-Pro", "description": "Stronger MiMo Pro tier for multimodal reasoning and coding-agent execution", "family": "mimo", "release_date": "2026-04-22", "last_updated": "2026-04-22", "knowledge_cutoff": "2024-12", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The weight repositories carry no LICENSE file; the MIT grant rests on the official open-sourcing announcement covering both V2.5 models including Base. The missing file caps clarity below a pinned-license grant.", "license_evidence": [ { "license_id": "MIT", "scope": "MiMo-V2.5-Pro and Base weights as declared in the official open-sourcing announcement", "kind": "web_terms", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced", "verified_at": "2026-09-18" } ], "access_boundary": "This record represents the April 2026 MiMo-V2.5-Pro release at Hugging Face commit 21d1ecfecd7bd70f31be25ca49d7edd21f003659 (last modified 2026-07-09), a distinct 1.02T/42B line from MiMo-V2.5, together with its API Platform serving. The Base sibling, the FP4-DFlash serving checkpoint, the UltraSpeed experience mode, later MiMo generations, and community quantizations remain separate boundaries.", "strengths": [ "Ungated trillion-parameter checkpoint with Base sibling, all MIT-claimed with commercial use allowed", "First-party API Platform and AI Studio serving for the exact model ID" ], "tradeoffs": [ "No LICENSE file ships beside the weights; the grant rests on announcement prose and tags", "1T total parameters need multi-node serving for self-operation" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 5, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 6.98 }, "evidence": [ { "kind": "web", "label": "Official model card and checkpoint", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official MiMo-V2.5 open-sourcing announcement with the MIT grant", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official MiMo-V2.5-Pro launch page (April 2026)", "url": "https://mimo.xiaomi.com/mimo-v2-5-pro/", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/xiaomi/mimo-v2.5-pro.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-xiaomi-mimo-v2-6-flash", "source_id": "xiaomi/mimo-v2.6-flash", "name": "MiMo-V2.6-Flash", "developer": "Xiaomi", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "description": "Xiaomi's MiMo-V2.6-Flash, a 309-billion-parameter sparse-MoE omnimodal model with 15B active parameters and 1M context, released September 22, 2026, as MIT open weights with a first-party API. models.dev reports open weights as false, which contradicts the first-party announcement.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiMo-V2.6-Flash", "description": "MiMo Flash model for multimodal coding agents and long-context automation", "family": "mimo", "release_date": "2026-09-22", "last_updated": "2026-09-22", "knowledge_cutoff": "2026-09-22", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The Hugging Face model card declares MIT and Xiaomi's V2.6 announcement states the weights are fully open-sourced. The repository card metadata is the governing evidence; models.dev's open-weights flag is stale.", "license_evidence": [ { "license_id": "MIT", "scope": "XiaomiMiMo/MiMo-V2.6-Flash-RL repository weights, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the MiMo-V2.6-Flash-RL release announced September 22, 2026, a distinct efficiency tier of the MiMo-V2.6 line. The MiMo-V2.6-Pro sibling, the reviewed V2 and V2.5 records, and the Pro UltraSpeed serving mode are separate boundaries.", "strengths": [ "Fully open weights under a declared MIT grant with a technical report and RL harnesses", "A distinct efficiency tier that outperforms the reviewed V2.5-Pro on agent benchmarks at a lower price" ], "tradeoffs": [ "A brand-new release with thin community runtime maturity", "The 309-billion-parameter dense-MoE checkpoint is heavy to self-host" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 6, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.3 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/xiaomi/mimo-v2.6-flash.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "verified_at": "2026-09-22" }, { "kind": "web", "label": "MiMo-V2.6 announcement", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2-6", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-xiaomi-mimo-v2-6-pro", "source_id": "xiaomi/mimo-v2.6-pro", "name": "MiMo-V2.6-Pro", "developer": "Xiaomi", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "description": "Xiaomi's MiMo-V2.6-Pro, the flagship sparse-MoE omnimodal model of the V2.6 line with 1.02T total and 42B active parameters and 1M context, released September 22, 2026, as MIT open weights with a first-party API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "MiMo-V2.6-Pro", "description": "Stronger MiMo Pro tier for multimodal reasoning and coding-agent execution", "family": "mimo", "release_date": "2026-09-22", "last_updated": "2026-09-22", "knowledge_cutoff": "2026-09-22", "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 1048576, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The Hugging Face model card declares MIT and Xiaomi's V2.6 announcement states the weights are fully open-sourced. models.dev's open-weights flag is stale.", "license_evidence": [ { "license_id": "MIT", "scope": "XiaomiMiMo/MiMo-V2.6-Pro-RL repository weights, Safetensors checkpoint, tokenizer, and chat template", "kind": "web_terms", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "verified_at": "2026-09-22" } ], "access_boundary": "This record represents the MiMo-V2.6-Pro-RL release announced September 22, 2026, the flagship tier of the MiMo-V2.6 line. The MiMo-V2.6-Flash sibling, the reviewed V2 and V2.5 records, and the Pro UltraSpeed serving mode (excluded as an alias) are separate boundaries.", "strengths": [ "Fully open weights under a declared MIT grant with a technical report and RL harnesses", "A distinct flagship tier marketed as the most powerful open-source model in its class" ], "tradeoffs": [ "The 1.02-trillion-parameter MoE is a serious self-hosting burden", "Evaluations are vendor-reported and the release is days old with immature community verification" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 9, "deployment_portability": 5, "serving_reach": 7, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.1 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/1bc95312f70441e339d5aa010ad656b9c7706952/models/xiaomi/mimo-v2.6-pro.toml", "verified_at": "2026-09-22" }, { "kind": "web", "label": "Official model page, model card, or weight repository", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "verified_at": "2026-09-22" }, { "kind": "web", "label": "MiMo-V2.6 announcement", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2-6", "verified_at": "2026-09-22" } ], "metadata_verified_at": "2026-09-22", "verified_at": "2026-09-22" }, { "id": "model-zhipuai-glm-4-5", "source_id": "zhipuai/glm-4.5", "name": "GLM-4.5", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.5", "description": "A developer-published hybrid-reasoning language-model release whose downloadable weights are declared MIT-licensed in the model card rather than in a repository license file.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.5", "description": "Hybrid-reasoning GLM release that made the 4.5 line broadly useful", "family": "glm", "release_date": "2025-07-28", "last_updated": "2025-07-28", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 98304 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.5" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The model card declares the GLM-4.5 weights released under the MIT open-source license for commercial use and secondary development, and the Hugging Face repository carries no separate license file. The companion GitHub code repository is licensed under Apache-2.0, which governs the published inference and fine-tuning code rather than the weights.", "license_evidence": [ { "license_id": "MIT", "scope": "zai-org/GLM-4.5 model weights, tokenizer, and chat template as declared in the pinned model card", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.5/blob/cbb2c7cfb52fa128a9660cb1a7a78e017899e115/README.md", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the zai-org/GLM-4.5 BF16 instruction release at commit cbb2c7cfb52fa128a9660cb1a7a78e017899e115 (last modified 2025-08-11). GLM-4.5-Air, the Base variants, the FP8 repositories, the Z.ai and Zhipu AI API services, and downstream quantizations remain separate boundaries.", "strengths": [ "The model card documents Transformers, vLLM, and SGLang serving paths with tested GPU configurations, and Z.ai's API documentation names glm-4.5 as a served model", "Developer-published Safetensors shards, tokenizer, and chat template, with Air, Base, and FP8 variants published as separate repositories" ], "tradeoffs": [ "The MIT declaration lives only in model-card metadata, and the GitHub code repository carries a different license (Apache-2.0)", "Neither the model card nor the API documentation states a release date, versioning policy, or deprecation posture" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.4 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and deployment guidance", "url": "https://huggingface.co/zai-org/GLM-4.5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Z.ai API documentation for the glm-4.5 model", "url": "https://docs.z.ai/guides/llm/glm-4.5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/zhipuai/glm-4.5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-zhipuai-glm-4-5-air", "source_id": "zhipuai/glm-4.5-air", "name": "GLM-4.5-Air", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.5-Air", "description": "An ungated 110B lightweight text model with a 131,072-token context, declared MIT on the model card and served on the Z.ai API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.5-Air", "description": "Lighter GLM-4.5 variant for fast coding assistance and cheaper agents", "family": "glm-air", "release_date": "2025-07-28", "last_updated": "2025-07-28", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 131072, "input": null, "output": 98304 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.5-Air" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries a frontmatter `license: mit` declaration and an explicit card statement that the models are released under the MIT open-source license and can be used commercially and for secondary development. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.5-Air/blob/a24ceef6ce4f3536971efe9b778bdaa1bab18daa/README.md", "verified_at": "2026-09-18", "scope": "Weights, tokenizer, and chat template in the official zai-org repository, as declared in the pinned model card frontmatter" } ], "access_boundary": "This record represents the zai-org/GLM-4.5-Air release of July 2025 at Hugging Face commit a24ceef6ce4f3536971efe9b778bdaa1bab18daa. The GLM-4.5 flagship, the vision lines, the GLM-4.6 and later generations, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated BF16 checkpoint with tokenizer and chat template in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the MIT grant rests on the frontmatter declaration", "No versioning or deprecation policy is published for the line", "The dedicated GLM-4.5-Air API documentation page is not served" ], "score_profile": "model_access", "score": { "license_clarity": 7, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.26 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-4.5-Air weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-4.5-Air", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-4.5 API documentation", "url": "https://docs.z.ai/guides/llm/glm-4.5", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-4.5-air.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-4-5v", "source_id": "zhipuai/glm-4.5v", "name": "GLM-4.5V", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.5V", "description": "An ungated 107B multimodal vision reasoning model handling text, image, and video input, declared MIT on the model card and served on the Z.ai API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.5V", "description": "GLM vision model for visual reasoning, documents, and multimodal agents", "family": "glm", "release_date": "2025-08-11", "last_updated": "2025-08-11", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 64000, "input": null, "output": 16384 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.5V" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries a frontmatter `license: mit` declaration and no grant prose in the card body. The grant rests on that declaration pointing at the standard MIT text, following the family posture of the reviewed GLM-5 release. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/README.md", "verified_at": "2026-09-18", "scope": "Weights, tokenizer, and chat template in the official zai-org repository, as declared in the pinned model card frontmatter" } ], "access_boundary": "This record represents the zai-org/GLM-4.5V release of August 2025 at Hugging Face commit ed47433b37111465ec527affaaddceff371bca04. The GLM-4.5-Air base model, the text-only lines, the GLM-4.6V and later vision generations, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated BF16 checkpoint with tokenizer and chat template in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the MIT grant rests on the frontmatter declaration", "No versioning or deprecation policy is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.18 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-4.5V weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-4.5V", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-4.5V API documentation", "url": "https://docs.z.ai/guides/vlm/glm-4.5v", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-4.5v.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-4-6", "source_id": "zhipuai/glm-4.6", "name": "GLM-4.6", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.6", "description": "An ungated 356B text model with a 204,800-token context, declared MIT on the model card and served on the Z.ai API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.6", "description": "Late GLM-4 workhorse for coding agents, reasoning, and structured tasks", "family": "glm", "release_date": "2025-09-30", "last_updated": "2025-09-30", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 204800, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.6" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries a frontmatter `license: mit` declaration and no grant prose in the card body. The grant rests on that declaration pointing at the standard MIT text, following the family posture of the reviewed GLM-5 release. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.6/blob/be72194883d968d7923a07e2f61681ea9a2826d1/README.md", "verified_at": "2026-09-18", "scope": "Weights, tokenizer, and chat template in the official zai-org repository, as declared in the pinned model card frontmatter" } ], "access_boundary": "This record represents the zai-org/GLM-4.6 release of September 2025 at Hugging Face commit be72194883d968d7923a07e2f61681ea9a2826d1. The GLM-4.5 generation, the vision lines, the GLM-4.7 and later generations, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated BF16 checkpoint with tokenizer and chat template in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the MIT grant rests on the frontmatter declaration", "No versioning or deprecation policy is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.18 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-4.6 weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-4.6", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-4.6 API documentation", "url": "https://docs.z.ai/guides/llm/glm-4.6", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-4.6.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-4-6v", "source_id": "zhipuai/glm-4.6v", "name": "GLM-4.6V", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.6V", "description": "An ungated 107B multimodal vision reasoning model handling text, image, and video input, declared MIT on the model card and served on the Z.ai API.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.6V", "description": "GLM vision model for visual reasoning, documents, and multimodal agents", "family": "glm", "release_date": "2025-12-08", "last_updated": "2025-12-08", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.6V" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries a frontmatter `license: mit` declaration and no grant prose in the card body. The grant rests on that declaration pointing at the standard MIT text, following the family posture of the reviewed GLM-5 release. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.6V/blob/4e2d47eb0b41c5280d8294b17cef9e94fdcfff46/README.md", "verified_at": "2026-09-18", "scope": "Weights, tokenizer, and chat template in the official zai-org repository, as declared in the pinned model card frontmatter" } ], "access_boundary": "This record represents the zai-org/GLM-4.6V release of December 2025 at Hugging Face commit 4e2d47eb0b41c5280d8294b17cef9e94fdcfff46. The 4.6V-Flash sibling, the text-only lines, the GLM-4.5V vision generation, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated BF16 checkpoint with tokenizer and chat template in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the MIT grant rests on the frontmatter declaration", "No versioning or deprecation policy is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.18 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-4.6V weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-4.6V", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-4.6V API documentation", "url": "https://docs.z.ai/guides/vlm/glm-4.6v", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-4.6v.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-4-6v-flash", "source_id": "zhipuai/glm-4.6v-flash", "name": "GLM-4.6V-Flash", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash", "description": "An ungated 9B multimodal vision reasoning model handling text, image, and video input, declared MIT on the model card and served on the Z.ai API free tier.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.6V-Flash", "description": "Lightweight GLM vision model for visual reasoning, documents, and multimodal agents", "family": "glm", "release_date": "2025-12-08", "last_updated": "2025-12-08", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 128000, "input": null, "output": 32768 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries a frontmatter `license: mit` declaration and no grant prose in the card body. The grant rests on that declaration pointing at the standard MIT text, following the family posture of the reviewed GLM-5 release. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash/blob/411bb4d77144a3f03accbf4b780f5acb8b7cde4e/README.md", "verified_at": "2026-09-18", "scope": "Weights, tokenizer, and chat template in the official zai-org repository, as declared in the pinned model card frontmatter" } ], "access_boundary": "This record represents the zai-org/GLM-4.6V-Flash release of December 2025 at Hugging Face commit 411bb4d77144a3f03accbf4b780f5acb8b7cde4e. The 4.6V flagship sibling, the text-only lines, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated BF16 checkpoint with tokenizer and chat template in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the MIT grant rests on the frontmatter declaration", "No versioning or deprecation policy is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.22 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-4.6V-Flash weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-4.6V API documentation", "url": "https://docs.z.ai/guides/vlm/glm-4.6v", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-4.6v-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-4-7", "source_id": "zhipuai/glm-4.7", "name": "GLM-4.7", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.7", "description": "An ungated 355B text model with a 204,800-token context, declared MIT on the model card and served on the Z.ai API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.7", "description": "Mature GLM model for dependable coding, reasoning, and structured agent tasks", "family": "glm", "release_date": "2025-12-22", "last_updated": "2025-12-22", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 204800, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.7" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries a frontmatter `license: mit` declaration and no grant prose in the card body. The grant rests on that declaration pointing at the standard MIT text, following the family posture of the reviewed GLM-5 release. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.7/blob/602d01efcdd332c5238ca4bcede555defbe83eb7/README.md", "verified_at": "2026-09-18", "scope": "Weights, tokenizer, and chat template in the official zai-org repository, as declared in the pinned model card frontmatter" } ], "access_boundary": "This record represents the zai-org/GLM-4.7 release of December 2025 at Hugging Face commit 602d01efcdd332c5238ca4bcede555defbe83eb7. The GLM-4.6 generation, the Flash and FlashX tiers, the vision lines, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated BF16 checkpoint with tokenizer and chat template in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the MIT grant rests on the frontmatter declaration", "No versioning or deprecation policy is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.04 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-4.7 weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-4.7", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-4.7 API documentation", "url": "https://docs.z.ai/guides/llm/glm-4.7", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-4.7.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-4-7-flash", "source_id": "zhipuai/glm-4.7-flash", "name": "GLM-4.7-Flash", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash", "description": "An ungated 30B text model with a 200,000-token context, declared MIT on the model card and served on the Z.ai API free tier.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-4.7-Flash", "description": "Budget GLM lane for fast coding help, routing, and everyday automation", "family": "glm-flash", "release_date": "2026-01-19", "last_updated": "2026-01-19", "knowledge_cutoff": "2025-04", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships no LICENSE file; the pinned model card carries a frontmatter `license: mit` declaration and no grant prose in the card body. The grant rests on that declaration pointing at the standard MIT text, following the family posture of the reviewed GLM-5 release. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash/blob/7dd20894a642a0aa287e9827cb1a1f7f91386b67/README.md", "verified_at": "2026-09-18", "scope": "Weights, tokenizer, and chat template in the official zai-org repository, as declared in the pinned model card frontmatter" } ], "access_boundary": "This record represents the zai-org/GLM-4.7-Flash release of January 2026 at Hugging Face commit 7dd20894a642a0aa287e9827cb1a1f7f91386b67. The GLM-4.7 flagship, the FlashX variant, the vision lines, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated BF16 checkpoint with tokenizer and chat template in the official repository", "The model card documents Transformers, vLLM, and SGLang deployment paths" ], "tradeoffs": [ "No LICENSE file ships with the weights; the MIT grant rests on the frontmatter declaration", "No versioning or deprecation policy is published for the line" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.18 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-4.7-Flash weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-4.7 API documentation", "url": "https://docs.z.ai/guides/llm/glm-4.7", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-4.7-flash.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-5", "source_id": "zhipuai/glm-5", "name": "GLM-5", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-5", "description": "A developer-published agentic-engineering language-model release with downloadable BF16 weights under an MIT declaration in the model-card metadata, served through the Z.ai API and documented third-party hosting.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GLM-5", "description": "General GLM flagship for coding, analysis, and tool-heavy engineering workflows", "family": "glm", "release_date": "2026-02-12", "last_updated": "2026-02-12", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 204800, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The zai-org/GLM-5 repository contains no license file; its model-card frontmatter declares MIT and the rendered page shows that declaration, matching the reviewed GLM-4.5 posture. The companion code repository is separately Apache-2.0 and governs code, not weights. The Z.ai API is governed separately by the platform terms.", "license_evidence": [ { "license_id": "MIT", "scope": "zai-org/GLM-5 model weights, tokenizer, and chat template as declared by the pinned model-card frontmatter; no repository license file", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/README.md", "verified_at": "2026-09-11" } ], "access_boundary": "This record represents the zai-org/GLM-5 BF16 instruction release. The GLM-5-FP8 repository, GLM-5.1, GLM-5.2, GLM-5.3 and their variants, the Turbo records, the GLM-4.x family, the Z.ai API and chat products, and downstream quantizations remain separate boundaries.", "strengths": [ "Developer-published BF16 Safetensors checkpoint with tokenizer and chat template plus documented vLLM, SGLang, KTransformers, Transformers, and xLLM serving paths", "First-party Z.ai API documentation names glm-5 as a served model, and Novita AI documents a third-party serverless path for the release" ], "tradeoffs": [ "The MIT declaration lives only in card frontmatter metadata with no license file and no grant prose in the card body", "Neither the model card nor the API documentation states a versioning policy or deprecation posture" ], "score_profile": "model_access", "score": { "license_clarity": 6, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 7, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.18 }, "evidence": [ { "kind": "web", "label": "Official model card, checkpoint, and deployment guidance", "url": "https://huggingface.co/zai-org/GLM-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Z.ai release blog for GLM-5", "url": "https://z.ai/blog/glm-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Official Z.ai API documentation for the glm-5 model", "url": "https://docs.z.ai/guides/llm/glm-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Novita AI model page documenting third-party serverless hosting", "url": "https://novita.ai/models/model-detail/zai-org-glm-5", "verified_at": "2026-09-11" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/zhipuai/glm-5.toml", "verified_at": "2026-09-11" } ], "metadata_verified_at": "2026-09-11", "verified_at": "2026-09-11" }, { "id": "model-zhipuai-glm-5-turbo", "source_id": "zhipuai/glm-5-turbo", "name": "GLM-5-Turbo", "developer": "Z.ai", "url": "https://docs.z.ai/guides/llm/glm-5-turbo", "description": "An API-served text foundation model optimized for agentic coding scenarios, available through the Z.ai API under the Z.AI terms of use, without downloadable model weights.", "model_type": "language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GLM-5-Turbo", "description": "Faster GLM-5 lane for coding agents that need lower latency", "family": "glm", "release_date": "2026-03-16", "last_updated": "2026-03-16", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-ZAI-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Z.ai provides GLM-5-Turbo as a managed service under its Terms of Use and Additional Terms for API Services, which reserve all rights in the models and prohibit extracting data, parameters, or code from the service. No model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-ZAI-API-Terms", "kind": "web_terms", "url": "https://docs.z.ai/legal-agreement/terms-of-use", "verified_at": "2026-09-18", "scope": "Access to GLM-5-Turbo through the Z.ai API under the Terms of Use and the Additional Terms for API Services" } ], "access_boundary": "This record represents the GLM-5-Turbo release served under the Z.ai API model name glm-5-turbo, launched March 2026. The GLM-5 and later open-weight releases, the multimodal 5V-Turbo sibling, the Z.ai API and chat products, and reseller surfaces remain separate service or product boundaries.", "strengths": [ "Z.ai publishes a dedicated API documentation page with the model identity, context, and benchmarks", "The ZClawBench optimization target and OpenClaw integration are documented first-party" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Z.ai publishes no versioning, deprecation, or retirement posture for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.88 }, "evidence": [ { "kind": "web", "label": "Official GLM-5-Turbo API documentation", "url": "https://docs.z.ai/guides/llm/glm-5-turbo", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.AI terms of use and API services terms", "url": "https://docs.z.ai/legal-agreement/terms-of-use", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-5-turbo.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-5-1", "source_id": "zhipuai/glm-5.1", "name": "GLM-5.1", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-5.1", "description": "An ungated 754B text model with a 200,000-token context, released under a standard MIT license file and served on the Z.ai API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-5.1", "description": "Strong GLM coding model for agentic engineering, terminals, and repository generation", "family": "glm", "release_date": "2026-04-07", "last_updated": "2026-04-07", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5.1" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships a standard MIT license text dated 2026, granting permission to use, copy, modify, and distribute the software without restriction, with the card frontmatter also declaring MIT. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/LICENSE", "verified_at": "2026-09-18", "scope": "GLM-5.1 weights, tokenizer, and bundled model code in the official zai-org repository" } ], "access_boundary": "This record represents the zai-org/GLM-5.1 release of April 2026 at Hugging Face commit 26e1bd6e011feb778d25ae34b09b07074139d92d. GLM-5, GLM-5.2, and later releases, the Turbo and vision lines, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated weights with a standard MIT license file shipped in the repository", "The model card documents SGLang, vLLM, xLLM, Transformers, and KTransformers serving paths" ], "tradeoffs": [ "No versioning or deprecation policy is published for the line", "Third-party hosting listings are not verified host documentation" ], "score_profile": "model_access", "score": { "license_clarity": 8, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.48 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-5.1 weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-5.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-5.1 API documentation", "url": "https://docs.z.ai/guides/llm/glm-5.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GLM-5.1 announcement", "url": "https://z.ai/blog/glm-5.1", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-5.1.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-5-2", "source_id": "zhipuai/glm-5.2", "name": "GLM-5.2", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-5.2", "description": "An ungated 753B text model with a 1,000,000-token context, released under a standard MIT license file with explicit open-source prose and served on the Z.ai API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-5.2", "description": "Open flagship GLM for long-horizon coding agents and million-token context work", "family": "glm", "release_date": "2026-06-13", "last_updated": "2026-06-13", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5.2" } ] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The repository ships a standard MIT license text dated 2026, the card frontmatter declares MIT, and the card body states the model is pure open source with no regional limits. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/LICENSE", "verified_at": "2026-09-18", "scope": "GLM-5.2 weights, tokenizer, and bundled model code in the official zai-org repository" } ], "access_boundary": "This record represents the zai-org/GLM-5.2 release of June 2026 at Hugging Face commit cf457fa734ab149ffef225f80893eb38c6ff5cdc, with the newer zai-org/GLM-5.3-BF16 named as successor. GLM-5 and GLM-5.1, the Turbo and vision lines, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated weights with a standard MIT license file and explicit open-source grant prose", "The card names the successor release and a migration guide is published" ], "tradeoffs": [ "No versioning or deprecation policy beyond the successor pointer is published", "Third-party hosting listings are not verified host documentation" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 7, "documentation_provenance": 8, "overall": 7.82 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-5.2 weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-5.2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.ai GLM-5.2 API documentation", "url": "https://docs.z.ai/guides/llm/glm-5.2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GLM-5.2 announcement", "url": "https://z.ai/blog/glm-5.2", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-5.2.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-5-3", "source_id": "zhipuai/glm-5.3", "name": "GLM-5.3", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-5.3", "description": "A developer-published agentic-coding language-model release with downloadable FP8 weights under the custom GLM-5.3 License carrying a large-operator model-as-a-service security-review condition, served through the Z.ai API.", "model_type": "language_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-5.3", "description": "Flagship GLM model for long-horizon coding, agents, and complex project delivery", "family": "glm", "release_date": "2026-08-14", "last_updated": "2026-08-14", "knowledge_cutoff": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "capabilities": { "attachment": false, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": true, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-GLM-5.3" ], "source_model": "source_available", "license_review_status": "verified", "license_note": "The zai-org/GLM-5.3 repository ships a real license file granting use, copying, modification, distribution, deployment, and fine-tuning, but requires any licensee operating a model-as-a-service business above 10 billion US dollars in aggregate affiliate revenue over any consecutive twelve months to pass a Z.AI security review before any commercial use. That gate breaks from the MIT posture of GLM-5, so the release is source-available rather than open source. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "LicenseRef-GLM-5.3", "scope": "zai-org/GLM-5.3 model weights, parameters, configuration files, inference and training code, and associated documentation; does not cover the hosted glm-5.3 API service", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the zai-org/GLM-5.3 FP8 instruction release. The BF16 sibling repository, GLM-5, GLM-5.1, GLM-5.2, GLM-5.3-Flash and its BF16 sibling, Turbo records, the GLM-4.x family, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Developer-published 141-shard FP8 checkpoint plus BF16 sibling, tokenizer, and chat template, with documented SGLang, vLLM, TokenSpeed, Transformers, and KTransformers serving paths", "First-party Z.ai API documentation names glm-5.3 as a served model with 1M context" ], "tradeoffs": [ "The custom GLM-5.3 License gates commercial use by large model-as-a-service operators behind a Z.AI security review, unlike the MIT posture of GLM-5", "Neither the model card nor the API documentation states a versioning policy or deprecation posture" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.0 }, "evidence": [ { "kind": "web", "label": "Official zai-org/GLM-5.3 weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-5.3", "verified_at": "2026-09-12" }, { "kind": "web", "label": "GLM-5.3 License file with model-as-a-service security-review clause", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official Z.ai API documentation for the glm-5.3 model", "url": "https://docs.z.ai/guides/llm/glm-5.3", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/zhipuai/glm-5.3.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12" }, { "id": "model-zhipuai-glm-5-3-flash", "source_id": "zhipuai/glm-5.3-flash", "name": "GLM-5.3-Flash", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash", "description": "A developer-published natively multimodal MoE release with downloadable MIT-licensed weights for efficient coding and long-horizon agents, served through the Z.ai API and documented third-party hosting.", "model_type": "multimodal_language_model", "distribution_modes": [ "downloadable_weights", "developer_api", "third_party_hosting" ], "source_metadata": { "name": "GLM-5.3-Flash", "description": "Native multimodal GLM model for efficient coding and long-horizon agent tasks", "family": "glm", "release_date": "2026-08-26", "last_updated": "2026-08-26", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": true, "temperature": true }, "limits": { "context": 1000000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "MIT" ], "source_model": "open_source", "license_review_status": "verified", "license_note": "The zai-org/GLM-5.3-Flash repository ships a real MIT license file and declares MIT in card frontmatter — stronger evidence than the reviewed GLM-5 frontmatter-only MIT posture, with identical MIT conclusion. The companion code repository is separately Apache-2.0 and governs code, not weights. The Z.ai API is governed separately by platform terms.", "license_evidence": [ { "license_id": "MIT", "scope": "zai-org/GLM-5.3-Flash model weights and associated documentation files under the repository MIT license; does not cover the hosted glm-5.3-flash API service", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/main/LICENSE", "verified_at": "2026-09-12" } ], "access_boundary": "This record represents the zai-org/GLM-5.3-Flash FP8 instruction release (62 Safetensors shards; weights released day-one with the API release). The BF16 sibling, GLM-5.3, GLM-5, GLM-5.1, GLM-5.2, Turbo records, the anonymous pre-release alias, the Z.ai API and coding-plan products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Developer-published 62-shard FP8 checkpoint plus MIT BF16 sibling, processor files, and chat template, with documented SGLang, vLLM, TokenSpeed, Transformers, and KTransformers paths", "First-party Z.ai API documentation for glm-5.3-flash plus fetch-verified Novita serverless hosting" ], "tradeoffs": [ "No versioning policy or deprecation posture is stated in the card or API documentation", "The card's fine-tuning guidance points at a shared guide rather than Flash-specific documentation" ], "score_profile": "model_access", "score": { "license_clarity": 9, "artifact_availability": 8, "deployment_portability": 8, "serving_reach": 8, "lifecycle_transparency": 7, "documentation_provenance": 9, "overall": 8.24 }, "evidence": [ { "kind": "web", "label": "Official model card, weights, and deployment guidance", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash", "verified_at": "2026-09-12" }, { "kind": "web", "label": "MIT license file for the GLM-5.3-Flash repository", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/main/LICENSE", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Official Z.ai API documentation for the glm-5.3-flash model", "url": "https://docs.z.ai/guides/llm/glm-5.3-flash", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Novita AI model page documenting third-party serverless hosting", "url": "https://novita.ai/models/model-detail/zai-org-glm-5.3-flash", "verified_at": "2026-09-12" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f57a900926160a597e96b62dab5de44732e95e35/models/zhipuai/glm-5.3-flash.toml", "verified_at": "2026-09-12" } ], "metadata_verified_at": "2026-09-12", "verified_at": "2026-09-12" }, { "id": "model-zhipuai-glm-5v-turbo", "source_id": "zhipuai/glm-5v-turbo", "name": "GLM-5V-Turbo", "developer": "Z.ai", "url": "https://docs.z.ai/guides/vlm/glm-5v-turbo", "description": "An API-served multimodal coding foundation model handling text, image, video, and document input, available through the Z.ai API under the Z.AI terms of use, without downloadable model weights.", "model_type": "multimodal_language_model", "distribution_modes": [ "developer_api" ], "source_metadata": { "name": "GLM-5V-Turbo", "description": "Fast GLM vision model for screenshots, documents, and multimodal agent tasks", "family": "glm", "release_date": "2026-04-01", "last_updated": "2026-04-01", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "capabilities": { "attachment": true, "reasoning": true, "tool_call": true, "structured_output": null, "temperature": true }, "limits": { "context": 200000, "input": null, "output": 131072 }, "reported_open_weights": false, "reported_license": null, "links": [], "weights": [] }, "licenses": [ "LicenseRef-ZAI-API-Terms" ], "source_model": "proprietary", "license_review_status": "verified", "license_note": "Z.ai provides GLM-5V-Turbo as a managed service under its Terms of Use and Additional Terms for API Services, which reserve all rights in the models and prohibit extracting data, parameters, or code from the service. No model weights are licensed for self-operation.", "license_evidence": [ { "license_id": "LicenseRef-ZAI-API-Terms", "kind": "web_terms", "url": "https://docs.z.ai/legal-agreement/terms-of-use", "verified_at": "2026-09-18", "scope": "Access to GLM-5V-Turbo through the Z.ai API under the Terms of Use and the Additional Terms for API Services" } ], "access_boundary": "This record represents the GLM-5V-Turbo release served under the Z.ai API model name glm-5v-turbo, launched April 2026 and built on the GLM-5-Turbo language base. The text-only Turbo sibling, the GLM-5 and later open-weight releases, the Z.ai API and chat products, and reseller surfaces remain separate service or product boundaries.", "strengths": [ "Z.ai publishes a dedicated API documentation page and a first-party technical report", "The report documents the multimodal agentic architecture and its language base" ], "tradeoffs": [ "No developer-published weights are available for self-operation", "Z.ai publishes no versioning, deprecation, or retirement posture for the line" ], "score_profile": "model_access", "score": { "license_clarity": 4, "artifact_availability": 1, "deployment_portability": 2, "serving_reach": 6, "lifecycle_transparency": 5, "documentation_provenance": 7, "overall": 3.88 }, "evidence": [ { "kind": "web", "label": "Official GLM-5V-Turbo API documentation", "url": "https://docs.z.ai/guides/vlm/glm-5v-turbo", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official Z.AI terms of use and API services terms", "url": "https://docs.z.ai/legal-agreement/terms-of-use", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Official GLM-5V-Turbo technical report", "url": "https://arxiv.org/abs/2604.26752", "verified_at": "2026-09-18" }, { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/9df518aae587f231d955a3f849306082c543aa4d/models/zhipuai/glm-5v-turbo.toml", "verified_at": "2026-09-18" } ], "metadata_verified_at": "2026-09-18", "verified_at": "2026-09-18" }, { "id": "model-zhipuai-glm-image", "source_id": "zhipuai/glm-image", "name": "GLM-Image", "developer": "Z.ai", "url": "https://huggingface.co/zai-org/GLM-Image", "description": "Z.ai's GLM-Image, a hybrid autoregressive plus diffusion-decoder image generation and editing model, published as ungated Safetensors weights declared MIT with Apache-2.0 tokenizer and vision components, and served on the Z.ai API as glm-image.", "model_type": "image_generation_model", "distribution_modes": [ "downloadable_weights", "developer_api" ], "source_metadata": { "name": "GLM-Image", "description": "GLM-Image is an image generation model adopts a hybrid autoregressive + diffusion decoder architecture. In general image generation quality, GLM‑Image aligns with mainstream latent diffusion approaches, but it shows significant advantages in text-rendering and knowledge‑intensive generation scenarios. It performs especially well in tasks requiring precise semantic understanding and complex information expression, while maintaining strong capabilities in high‑fidelity and fine‑grained detail generation. In addition to text‑to‑image generation, GLM‑Image also supports a rich set of image‑to‑image tasks including image editing, style transfer, identity‑preserving generation, and multi‑subject consistency.", "family": null, "release_date": "2026-01-19", "last_updated": "2026-01-19", "knowledge_cutoff": null, "modalities": { "input": [ "text", "image" ], "output": [ "image" ] }, "capabilities": { "attachment": true, "reasoning": false, "tool_call": false, "structured_output": null, "temperature": false }, "limits": { "context": 10240, "input": null, "output": 0 }, "reported_open_weights": true, "reported_license": "MIT", "links": [], "weights": [ { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-Image" } ] }, "licenses": [ "MIT", "Apache-2.0" ], "source_model": "mixed_open_source", "license_review_status": "verified", "license_note": "The model card states in prose that the overall GLM-Image model is released under the MIT License and that the VQ tokenizer and vision-encoder weights taken from X-Omni remain under Apache-2.0, so both licences govern parts of the weights. The Hugging Face repository ships no LICENSE file, so the MIT grant rests on the card's statement; the GitHub repository carries an Apache-2.0 LICENSE file dated 2026 for its code, which the card does not reconcile with the MIT statement. All named licences are open and no field-of-use restriction appears, so the release is classified as mixed open licenses. The Z.ai API serving glm-image is governed separately by the Z.ai Terms of Use and Additional Terms for API Services.", "license_evidence": [ { "license_id": "MIT", "scope": "Overall GLM-Image model weights in the official zai-org repository, as stated in the model card License section", "kind": "web_terms", "url": "https://huggingface.co/zai-org/GLM-Image/blob/main/README.md", "verified_at": "2026-10-06" }, { "license_id": "Apache-2.0", "scope": "GLM-Image GitHub repository code under its LICENSE file, and the VQ tokenizer and vision-encoder weights from X-Omni that the model card says remain under Apache-2.0", "kind": "web_terms", "url": "https://github.com/zai-org/GLM-Image/blob/main/LICENSE", "verified_at": "2026-10-06" } ], "access_boundary": "This record represents the zai-org/GLM-Image weight release announced January 14, 2026 (repository last modified 2026-01-15) together with the glm-image model served on the Z.ai API. The GLM-4 text and vision lines, the CogView models, the Z.ai API and chat products, ModelScope mirrors, and downstream quantizations and fine-tunes remain separate boundaries.", "strengths": [ "Ungated Safetensors weights with a documented diffusers pipeline and an SGLang pipeline, plus a published technical blog", "The same release is served on the first-party Z.ai API as glm-image, with the model identifier documented" ], "tradeoffs": [ "No MIT licence text ships with the weights, the card's MIT statement sits beside an Apache-2.0 repository licence file, and two components carry their own Apache-2.0 terms", "No versioning or deprecation policy is published for the model" ], "score_profile": "model_access", "score": { "license_clarity": 5, "artifact_availability": 9, "deployment_portability": 8, "serving_reach": 6, "lifecycle_transparency": 6, "documentation_provenance": 8, "overall": 7.0 }, "evidence": [ { "kind": "web", "label": "Pinned models.dev source metadata", "url": "https://github.com/anomalyco/models.dev/blob/f9f2057be3c620898b6eaaca91dbbf75176043ae/models/zhipuai/glm-image.toml", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official zai-org/GLM-Image weights repository and model card", "url": "https://huggingface.co/zai-org/GLM-Image", "verified_at": "2026-10-06" }, { "kind": "web", "label": "zai-org/GLM-Image GitHub repository with Apache-2.0 LICENSE file", "url": "https://github.com/zai-org/GLM-Image", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official GLM-Image announcement blog dated January 14, 2026", "url": "https://z.ai/blog/glm-image", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Official Z.ai GLM-Image API documentation", "url": "https://docs.z.ai/guides/image/glm-image", "verified_at": "2026-10-06" }, { "kind": "web", "label": "Z.ai Terms of Use and Additional Terms for API Services", "url": "https://docs.z.ai/legal-agreement/terms-of-use", "verified_at": "2026-10-06" } ], "metadata_verified_at": "2026-10-06", "verified_at": "2026-10-06" } ] }