Reflection Launches Beam: 501B-Parameter Open MoE Model

Reflection announced Beam, a text-only 501B-total / 23B-active MoE model for coding, agentic and scientific work, trained from scratch with full Apache 2.0 weights due this month. The team cites 23.8T pretraining tokens, a stable RL/OPD run on 10K GB300s with over 100M rollouts, and claimed 80.9 on SWE-bench Verified. It marks Reflection's arrival as a functional neolab in the US open-model ecosystem, though competitors like GLM 5.3, Kimi K3, Qwen 3.8 Max, and DeepSeek V4.1 Flash are generally ahead.