Reflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads

Reflection AI has introduced Beam, an open-weight sparse Mixture-of-Experts model featuring 501B parameters and 23B active parameters. Designed for coding and agentic workloads, Beam matches GLM-5.2 in reasoning while requiring 3 to 4x less inference compute. Apache 2.0 weights will be released in October 2026. This launch follows a funding round valuing the startup at $25 billion and a $6.3 billion agreement with SpaceX Corp. to utilize Nvidia GB300 NVL72 appliances for development.