Beam

Beam is Reflection’s first open-weight model, a text-only Mixture-of-Experts model with 501 billion total parameters and 23 billion active per token, built for coding, reasoning, and agentic workloads. Pretrained on 23.8 trillion tokens and refined through large-scale reinforcement learning, it combines strong software engineering and tool-use capabilities with efficient inference. Users can adjust its reasoning effort to balance response length, performance, and compute cost.