Reflection releases Beam, a 501B mixture-of-experts model for coding and agents

Reflection releases Beam, a text-to-text mixture-of-experts model with 501B total parameters, 23B active per token, and a context window of up to 1 million tokens. Reinforcement learning with longer rollouts and a length penalty aims to solve coding and agentic tasks with fewer tokens. Beam is available through Baseten Model APIs and Dedicated Inference.