Reflection open-sources Beam: 501B-param MoE with 156-page tech report
_dsevero · x · 2026-11-07
Reflection has released the weights for Beam along with a 156-page technical report and a safety report.
Key facts:
- Architecture: sparse MoE with 501.3B total / 23.4B active parameters, built for coding, reasoning, and agentic workloads; delivers frontier-level reasoning with 3-4x less inference compute than peers
- Pretraining: 23.8T curated tokens; 1M-token context via a novel two-stage midtraining phase
- RL scale: fully asynchronous RL campaign across 10.5K GB300 GPUs, generating 100M+ rollouts in 4 weeks — claimed as one of the largest RL runs by any open lab; capabilities kept improving with more RL compute with no plateau
- Serving: FP8/NVFP4 quantization and a 1.93B speculative decoding drafter (DFlash 2)
- Safety: refuses or pivots on 88.9% of harmful requests while complying with 81.5% of harmless ones
The report covers pretraining, midtraining, SFT, RL, and on-policy distillation plus infra, data, and research decisions; weights, playground, and API access are live.
Related event: Reflection AI Open-Sources Beam with 156-Page Technical Report(8 posts)→
More from Infra
- AI server makers Dell, HPE, Lenovo and Supermicro hike prices by up to 45%, even on signed deals — churchkey · 2026-11-22
- Where will the world store AI's exploding data? Oceans, space, DNA and glass — ChuckDBrooks · 2026-11-22
- Codex Code Mode calls get 10ms faster by reusing tool catalogs across cells — charliermarsh · 2026-11-22
- AI ecosystem guide 2026 maps 120 tools across 18 categories from training to agents — blaizedsouza · 2026-11-22
- Google AI members get free monthly Cloud credits: $10/mo for Pro, up to $100/mo for Ultra 20x — lxfater · 2026-11-22
- Inkling 2 mini's 1M-context KV cache dissected: only 8 of 48 layer visits keep full cache — teortaxesTex · 2026-11-22