ChefNMR2 is a novel AI system for rapid automated de novo small-molecule structure elucidation from 1D ¹H and ¹³C NMR spectra and molecular formulae. It uses an atomic flow-matching generative model trained on 2.52 million molecules, optionally augmented by an LLM reasoning agent (ChefNMR2-Agent). On five benchmarks totaling 2,031 molecules it achieved 82.4% top-1 accuracy (vs 51.9% for Claude Opus 5). The system is available to partners inside the Kosmos AI scientist platform; a public web app, MCP tool, code and paper were also released.
ChefNMR2 is a generative model that proposes small-molecule structures from molecular formulae and 1D ¹H and ¹³C NMR peak lists. Its inference code can be run locally, and a browser demo and MCP connection are also documented. Edison Scientific reports that, on five benchmarks totaling 2,031 molecules, ChefNMR2 ranked the reported structure first in 82.4% of cases. The related ChefNMR2-Agent combines model predictions with LLM-based chemical reasoning; Edison says it was used with Kosmos to investigate four previously unpublished natural products.
Be the first Cosign
Add it to your Products I use list. You can also share why you use it.
