Energy
Periodic Neon — 1T lab-trained model beats GPT-6 on XRD analysis
Periodic Labs introduced Periodic Neon, a 1-trillion-parameter model midtrained and RL-tuned on its own high-throughput lab data to interpret X-ray diffraction results. Deployed in Menlo Park labs that run 24/7, Neon outperforms Claude Fable 5.1 and GPT-6 Astra on that analysis task with relatively little compute. The company uses the loop of hypothesize, synthesize, and characterize to train scientific AI that then directs experiments toward new superconductors and magnets. It matters because it turns physical lab data into a closed loop for autonomous materials discovery.
periodic.com