Periodic Labs introduced Periodic Neon, a 1T-parameter AI model. It analyzes experimental results from physical labs. The model was post-trained on Periodic’s own lab data. It outperforms GPT-6 Astra on diffraction analysis. Neon uses midtraining and reinforcement learning on lab data. It deploys in Periodic’s Menlo Park labs running 24/7. The model embeds in lab equipment to guide experiments. Periodic breaks materials discovery into three phases. It forms a loop of hypothesis, synthesis prediction and result analysis. The company aims to train AI to direct scientific campaigns. It seeks to discover materials nobody yet knows how to make.
What changed
Scientific analysis of lab results relied on manual judgment and separate simulations.
What it unlocks
Automated interpretation of X-ray diffraction results in high-throughput labs.
- 1T parameters
- outperforms GPT-6 Astra on diffraction analysis
Sources