Reflection unveils Beam: 501B-param open MoE model with 23B active, trained on 23.8T tokens
bigblueboo · x · 2026-10-06
Reflection AI announced Beam, its first open-weight model, with full weights, technical report and model card due later this month.
- Architecture & training: sparse MoE with 501B total / 23B active parameters; pretrained from scratch on 23.8 trillion curated tokens
- Heavy RL investment: a 4-week high-compute RL run on 10.5K NVIDIA GB300 GPUs generated over 100 million rollouts
- Positioning: focused on coding, reasoning and agentic workloads; claims to advance the Western open-weight frontier, competitive with GLM 5.2 and approaching Qwen 3.8-Max on coding/agentic tasks, with inference efficiency as its edge over stronger open models like Kimi K3
- Currently in final red-teaming; early access sign-ups are open
Related event: Reflection AI Unveils Beam, Its First Open-Weight 501B-Parameter MoE Model(13 posts)→
More from Models
- Indie chatbot Auro V8 adds cross-session memory, claims self-iterating training loop — TheMoonMidas · 2026-10-06
- Dev warns: Pangram AI detector doesn't work and shouldn't be taken seriously — jdjohnson · 2026-10-06
- Reflection.ai Introduces Beam, a 501B Open-Weight Model — laurenthuberd · 2026-10-06
- Gemini 4 Argon tops Text Arena at 1525 pts, 20 pts above Claude Opus 4.6, at $8/MToken — arena · 2026-10-06
- Arena CEO: the moment you pay people to create benchmarks, they stop being real — arena · 2026-10-06
- OpenAI ships compaction in Responses API, sparking vendor lock-in debate among developers — pvncher · 2026-10-06