Periodic Labs, building labs that learn
15 hours ago
- Building high-throughput physical labs and large-scale simulations for discovering new materials, running 24/7 with enhanced throughput.
- Using AI to analyze experimental data and decide next steps, with custom training when frontier models fall short or are too costly.
- Periodic Neon, a 1T-parameter model, outperforms GPT-6 Astra and Claude Fable 5.1 on diffraction analysis using lab data and relatively little compute.
- Training Periodic Neon via midtraining and reinforcement learning on lab data to achieve frontier-level reasoning and deploy at scale.
- Breaking materials discovery into a three-phase loop: hypothesis, prediction of synthesis, and analysis of results, moving closer to a digital regime.
- Extending AI training to direct scientific campaigns, develop synthesis procedures, and decide which experiments to run.