Reflection AI debuts Beam: 501B-param MoE with 23B active, open weights this month
alexpolozov · x · 2026-10-06
Reflection AI announced Beam, its first open-weight model: a highly efficient agentic MoE with 501B total and 23B active parameters, trained end-to-end from scratch and claimed to advance the Western open frontier on coding and agentic tasks. Full weights release this month.
- Pretraining lead tolgab0 (ex-Gemini, joined Dec 2024) recalls the team was under 10 people at the start, building in-house evals, data materialization, scaling ladders, and large-scale data filtering pipelines from scratch
- The team stressed scientific rigor: every improvement was measured and validated, with compute efficiency and training stability steadily improving as they scaled
More from Models
- Developer slams Grok models as 'garbage', shares coding agent's repetitive trial-and-error log — feregri_no · 2026-11-20
- Drex Router cuts small-game build cost to $1.26 vs $13.44 on Claude Opus alone — testingcatalog · 2026-11-20
- OpenAI slammed for AI slop: 'Just hire photographers to RLHF imagegen for a day' — flowersslop · 2026-11-20
- Model Fight Card 16: Tenet models dominate jailbreak arena with 21/24 escapes — Dave_Maynor · 2026-11-20
- Coding agent index shows max effort costs 6x for fewer points; Gemini 4 Neon wins on value — Arindam_1729 · 2026-11-20
- Reddit: Claude Max for $29 Instead of $200 — How Do Resellers Profit? — mehdiweb · 2026-11-20