Reflection AI releases Beam, a 501B Apache-2.0 open-weight agentic coding model, on Hugging Face and AWS

Reflection AI's Beam is the latest in a dense run of large open-weight releases this week. It is a 501B-parameter model aimed squarely at agentic coding and long-horizon tasks, and it ships under the permissive Apache 2.0 license with more than 40 launch partners.

Distribution is broad from day one. The weights are on Hugging Face. AWS supports SageMaker JumpStart deployment, HyperPod Inference, and serving through SGLang or vLLM, plus Trainium for cost-sensitive inference.

The licensing is the differentiator. Meta's Muse Spark Open carries a custom Muse Model License with limited commercial terms, and Mistral Large 4's full weights don't arrive until October 31. Beam's Apache 2.0 terms impose almost no restrictions, which community commentators flagged as the most enterprise-friendly option among this week's open releases.

What to watch: independent coding benchmarks against Muse Spark Open and DeepSeek V4.1, and whether 501B is practical to self-host outside multi-GPU clusters.