Periodic Labs · 5× · naposledy 17. 9. 2026
Periodic Neon is a 1-trillion-parameter model developed by Periodic Labs, built on the Open Source Kimi-K2.6 model through scientific Intermediate Training (technical literature, code) and subsequent Reinforcement Learning with proprietary lab data. It is deployed internally in Periodic Labs' physical high-throughput laboratories to analyze experimental results—particularly complex X-ray diffraction (XRD) data from materials research (including superconductors, magnets, semiconductors). According to proprietary (not yet externally peer-reviewed) Benchmarks on "FrontierXRD," Neon outperforms both the unmodified base model and commercial models such as GPT-6 Astra and Claude Fable 5.1. Model weights, pricing, or public API are not available.