Reflection AI launches Beam, a 501B-parameter open-weight agentic model with 23B active, after one year
simonguozirui · x · 2026-10-06
Reflection AI introduced Beam, its first model trained end-to-end from scratch: a highly efficient open-weight agentic model with 501B total parameters and 23B active, claiming frontier reasoning efficiency and Western open-weight frontier advances on coding and agentic tasks. The pre-training team started from scratch with just 5 people last November; full weights drop this month.
More from Companies & People
- Polymarket puts 83% odds on Google having the best AI model by year-end — Polymarket · 2026-11-20
- Over 540 Google employees petition for an AI-free Google Search in K-12 schools — Polymarket · 2026-11-20
- YC-backed Antinuous launches: 100x more experiments, 1,000x cheaper science — ycombinator · 2026-11-20
- Adaption AI to host NYC gathering for AI researchers and leaders on Dec. 3 — hugo_larochelle · 2026-11-20
- Wired podcast: the outages, Anthropic's IPO money, and AI chips post-Jensen Huang — Wired AI · 2026-11-20
- Nous Research partners with Microsoft to bring Hermes Agent into Teams and Azure — NousResearch · 2026-11-20