Model · Language model

Beam

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.

Developer

Reflection AI

Distribution

Developer API

Artifact licensing

Proprietary access

License scope

Reviewed release artifacts and mandatory terms; not an assessment of training code or training data openness.

Licenses

Proprietary

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.

Open in the directory → Open official model page ↗

Editorial ratings appear in the directory beside the profile they belong to and are never compared across collections.