Reflection AI unveils Beam, a 501B open-weight MoE built to take on China's open models
Reflection AI announced Beam, a sparse mixture-of-experts model with 501B total and 23B active parameters, pretrained on 23.8T tokens and then reinforcement-trained with 100M+ rollouts on 10,500 Nvidia GB300 GPUs. The company reports 80.9 on SWE-Bench Verified, 77.2 on SWE-Bench Pro v2-Hard and 80.1 on Terminal Bench v2.1, and says Beam matches Z.ai's GLM-5.2 with roughly 3-4x less inference compute while still trailing Kimi K3 and DeepSeek V4.1 Flash on raw capability. Weights ship under Apache 2.0 later this month after red-teaming, with a 1M-token context and early access open now. For builders, it is the first US-made open-weight model at this scale in months, and a credible alternative to depending on Chinese open models for self-hosted agents.
