ChefNMR2 launches to identify small-molecule structures from NMR spectra

ChefNMR2 uses a molecular formula and proton and carbon nuclear magnetic resonance (NMR) peak lists to generate candidate structures. In the team's tests of 2,031 molecules, its top-ranked planar structure was correct 82.3% of the time, versus 52.0% for the best tested LLM. Researchers say its reasoning agent helped identify four novel natural products. The model does not predict stereochemistry. Its inference code is available, while the agent's broader release is still in development.