Reflection AI Open-Weights Beam, a 501B Apache 2.0 Model Hitting ~80% on Terminal-Bench 2.1 at 25K Tokens

Reflection AI released Beam on Oct 16 under Apache 2.0 — a 501B-total / 23B-active MoE for coding, agents, science and math. The company says Beam hits ~80% on Terminal-Bench 2.1 while spending ~25,000 tokens per attempt, versus GLM-5.2 at 80,000 tokens for similar scores. Training ran four weeks on 10,500 Nvidia GB300 GPUs with 100M+ RL rollouts that penalized unnecessary tokens; weights fit on an eight-GPU server in 8-bit or 4-bit.