Periodic Labs introduces Periodic Neon, a large AI model post-trained on its lab data for X-ray diffraction analysis

Periodic Labs/Autonomous materials lab

Periodic Labs introduced Periodic Neon, a 1-trillion-parameter AI model further trained, with additional training and reinforcement learning, on data from its own labs in Menlo Park. Its first job is analysing X-ray diffraction, the measurement that shows a material's crystal structure and so whether a synthesis made what was intended. Periodic says Neon beats Claude Fable 5.1 and GPT-6 Astra at this task, but published no numbers.

• Reading diffraction patterns has traditionally taken hours of expert judgement per sample.
• The labs run 24/7 in a loop: predict stable target materials, predict how to make them, measure, and feed results back to improve the predictions.
• While multi-day experiments run, the team retrains on existing experimental data instead of leaving compute idle.
• Periodic notes the limits of learning from physical experiments: they are slow, hard to scale and often give ambiguous results.