Periodic Labs' Neon took X-ray diffraction analysis from 2.7% to 55.3% by training inside its own physical lab

Latent Space (AINews)·high signalXBlueskyLinkedInCopy link

Periodic Labs mid-trained and post-trained a 1T-parameter model called Neon on roughly 1,300 H200s using months of fresh data generated by its own high-throughput materials labs, closing a model-experiment feedback loop rather than learning from a static corpus. Liam Fedus reported the system moved hard X-ray diffraction analysis from 2.7% to 55.3% across 134 samples, and the team claims it surpasses GPT-6 Astra on their analysis benchmark. The interesting structural claim for anyone building domain agents is that the proprietary bottleneck was physical experiment throughput, not compute or base model quality.