Reflection AI built Beam, a 501-billion-parameter open-weight MoE model (23B active) for coding and agent tasks, planning Apache 2.0 weight release later in October.
On Terminal Bench v2.1, Beam scored 80.1 vs GLM 5.2’s 81.0, Kimi K3’s 88.3, and DeepSeek V4.1 Flash’s 90.6. On DeepSWE v1.1 Beam scored 44.4 vs GLM 5.2’s 44.0.
Reflection says Beam scores comparably to GLM 5.2 on advanced reasoning benchmarks while using 3-4x less inference compute. Benchmark claims are self-reported and not independently verified.
Beam remains in final red-teaming. Reflection trained a separate safety model from the same checkpoint and combined via distillation.