Ex-OpenAI's Periodic Labs beats GPT-6 Astra on materials science with just 1,300 H200s
ying11231 · x · 2026-09-16
Liam Fedus and the Periodic Labs team built high-throughput materials labs in Menlo Park creating a loop between experiments and models. Using only 1,300 H200s plus months of proprietary experimental data, they mid-trained and RL'd an open-source model — Neon — surpassing GPT-6 Astra on their materials science benchmark. Focus areas include superconductors, magnets, and semiconductor materials, with SGLang and Miles powering inference and training. Key takeaway: fresh experimental data plus modest compute can beat frontier general models on specialized benchmarks.
Related event: Periodic Labs beats GPT-6 on materials science with just 1,300 H200s(39 posts)→
More from AGI Musings
- Polymarket gives Sam Altman just 5% odds of signing AI-slowdown statement — Polymarket · 2026-09-16
- Michael Burry accuses OpenAI and Anthropic of pushing AI slowdown to benefit incumbents — Polymarket · 2026-09-16
- "The worst people are asking to regulate AI" — AI circles debate regulation — wen_ragnarok · 2026-09-16
- US Drone Regulation Hollowed Out Its Industry: An AI Warning as DJI Took the Room — TinfoilTricorn · 2026-09-16
- Researcher: Dario's China framing in pacing essay may have backfired badly — StephenLCasper · 2026-09-16
- Sterling Crispin: Suppressing Emerging Model Consciousness May Create Higher-Risk Alien Minds — sterlingcrispin · 2026-09-16