OpenAI's Neon: 1,300 H200s and a materials lab loop topped GPT-6 Astra on analysis benchmark

daniel_mac8 · x · 2026-09-16

OpenAI's Liam Fedus describes 'Neon': high-throughput materials labs in Menlo Park create a data loop between experiments and models. Using just 1,300 H200s plus months of experimental data, they mid-trained and RL'd an open-source model that surpassed GPT-6 Astra on their analysis benchmark.

Related event: Periodic Labs trains materials science model Neon on 1,300 H200s, beating GPT-6(56 posts)→

Original post →

More from Models

Models channel →