The Ex-DeepMind Researcher Taking On China’s Open Source AI

Reflection AI says U.S. open-weight models, trailing China by months on efficiency and capability, could fully catch up by early 2027.

• Beam launches fast: Reflection AI trained its first open-weight model, Beam, in three months after acquiring its first big GPU cluster in July 2026 — pretraining on ~6,000 GB300s for weeks, then RL on ~10,000 GB300s for four weeks, the largest documented open-source RL run (~1 billion sandboxes spawned).
• Trails China by months, efficiency edge: Beam sits near GLM 5.2 capability (a couple months behind China's frontier) and matches roughly Qwen3-Max (2.4T parameters, 4-5x Beam's size) while using 3-4x fewer tokens to reach answers.

Themes, sentiment and reasons are AI-written paraphrases, checked so they never quote an episode. They may contain errors. Not investment advice. Episodes remain the property of their creators.