Reflection AI Releases Beam Weights With Post-Training Workflows

Reflection AI has released downloadable weights for Beam, its model for coding, reasoning and agentic tasks, alongside documentation and integrations for developers looking to customize it.

Beam has 501.3 billion total parameters, with 23.4 billion active, and supports a one-million-token context window. The text model is available in BF16, FP8 and NVFP4 formats under an Apache 2.0 license.

The release includes workflows for fine-tuning and reinforcement learning, along with tools to evaluate changes to model capabilities and safety. A separate speculative-decoding drafter model is also available. Self-distillation recipes are planned for a later release.

Reflection introduced Beam on Oct. 5. The model’s weights became available on Oct. 16, alongside an ecosystem of more than 40 partners and integrations.