Reflection releases Beam, most capable open-weight model outside China
reflection released beam, its first open-weight model, which matches glm 5.2 on key benchmarks with 3-4x less compute. this affects businesses using ai for coding and automated workflows.
reflection, an ai startup, has announced beam, its first freely available model. beam is a mixture-of-experts model that activates 23 billion of its 501 billion total parameters per token, keeping compute costs relatively low. it was trained with reinforcement learning on 10,500 nvidia gpus over four weeks. beam matches glm 5.2 on demanding reasoning tasks while using 3-4x less compute. the model is designed for coding, logical reasoning, and agentic tasks, and will ship later this month under the apache 2.0 license.