Beam 1.0 501B-A23B-FP8
Beam is Reflection’s first open-weight model: a text-only, 501-billion-parameter Mixture-of-Experts model with 23B active parameters per token and a one-million-token context window, built for coding, reasoning, and agentic workloads.
Model
Beam combines a sparse Mixture-of-Experts architecture with large-scale reinforcement learning focused on software engineering, tool use, and multi-step reasoning. Its reasoning effort parameter lets you balance response length and task performance: lower settings favor shorter responses, while higher settings allow more reasoning on demanding tasks.
Reflection trained Beam to improve capability while limiting unnecessary token generation. Beam achieves reasoning performance comparable to GLM 5.2 while using approximately 3–4× less estimated inference compute.
Full details are in Reflection’s announcement.
Dedicated Endpoint
DocsA Dedicated Endpoint is your own deployment of Beam 1.0 501B-A23B-FP8, on an inference stack that Modal tunes and autoscales. Bring fine-tuned weights if you have them. Compute is billed by the second, and only while it runs.
modal endpoint create --model reflection/Beam-1.0-501B-A23B-FP8