Reflection AI's Beam: a new 501B MoE open-weights coding model
Melvin Vivas · X post · 2026-10-17 · Open on X
Topics: Industry Trends & Job Market, LLM Fundamentals · Level: intermediate
Summary
The creator points to Beam, an agentic open model from Reflection AI. It is a mixture-of-experts model with 501B total parameters, of which 23B are active. Reflection says it was trained end-to-end from scratch, is very efficient at reasoning, and pushes the Western open-model frontier on coding and agentic tasks. The full weights are due to be released this month.
Key points
- Beam is an open-weights model built for coding and agentic tasks
- It has 501B total parameters but only 23B active per token, a mixture-of-experts design
- Reflection says it was trained end-to-end from scratch
- It's pitched as highly efficient at reasoning
- Reflection claims it advances the Western open frontier on coding and agentic benchmarks
- The full weights are promised for release this month
Resources mentioned
- Beam (Reflection AI) · tool · reflection.ai · free
An open-weights agentic coding model with 501B total and 23B active parameters. - Reflection AI (@reflection_ai) on X · person · x.com · free
The X account of Reflection AI, the lab behind the Beam open model.
Try this
- Watch for the full Beam weights release and try it on coding or agentic tasks
More in Industry Trends & Job Market
- Jensen Huang on Why Open Models Matter and Shouldn't Be Regulated Away
- DeepSeek-V4.1-Flash-1017 release: better benchmarks, shorter outputs
- Gemini 4 Argon now available in Devin
- Glimpse: A Visual Reasoning Model That Uses a Visual Scratch Space
- Grok Bot vs OpenAI in the consumer AI agent market
- Why Simpler Agent Bots Appeal to Non-Technical Users