Periodic Labs' Neon tops GPT-6 Astra using just 1,300 H200s and closed-loop science
BanghuaZ · x · 2026-09-16
Periodic Labs unveiled Neon, an open-source model for materials science trained with mid-training and RL on only 1,300 H200s plus months of proprietary lab data, surpassing GPT-6 Astra on its analysis benchmark.
Key points:
- The team built high-throughput materials labs in Menlo Park to close the loop between experiments and models: labs generate fresh data, models learn from it, then suggest what to try next — targeting hard problems like superconductors, magnets, and semiconductor materials.
- The training stack extends the open-source SGLang and Miles to trillion-parameter-scale scientific RL, with more efficient training, lower memory use, and 2.5x faster inference; the improvements were contributed back to both projects.
Related event: Periodic Labs beats GPT-6 on materials science with just 1,300 H200s(41 posts)→
More from Companies & People
- Dario Amodei Shows Up at Dreamforce, Day After 'AI Will Kill Us' Jabs — inductionheads · 2026-09-16
- Jensen Huang Goes Unusually Bullish on Open Source: 'Every Company Will Become an AI Company' — kimmonismus · 2026-09-16
- OpenAI signs new OneGov deal with GSA to supply AI to all federal government employees — Felipe_Millon · 2026-09-16
- TotalEnergies and Mistral strike €100M+ AI deal to build frontier models for oil exploration — arthurmensch · 2026-09-16
- Benioff coaches Jensen Huang on walking and talking at Dreamforce, drawing laughs — TheTuringPost · 2026-09-16
- Jensen Huang: 'the end of software' is nonsense, and AI destroying jobs is untrue — kimmonismus · 2026-09-16