LaunchAI ModelsOctober 5, 2026

Reflection AI unveils Beam, a 501B open-weight MoE model

Read original source →reflection.ai

Beam is a sparse MoE with 501B total and 23B active parameters, trained from scratch on 23.8T tokens in under four weeks on 6,144 NVIDIA GB300 GPUs. Reflection claims 80.9 on SWE-bench Verified and 3-4x the inference efficiency of GLM 5.2; full Apache 2.0 weights are due this month.

How this story unfolded

1 day · 4 reports · 7 community posts · 11 of 12 shown

  1. Oct 5
  2. Oct 6

More stories today

Open the live feed