OpenAI launched MedChemBench, a benchmark of 99 medicinal-chemistry problems to assess AI models’ drug-discovery reasoning.

On Nov. 20, 2026, OpenAI introduced MedChemBench, a 99-problem test of whether AI can interpret molecular structures and experimental results, compare compounds, and predict their properties. The problems draw on 117 papers and patents and cover 761 distinct molecular structures. At the xhigh reasoning setting, GPT-6.2 Astra scored 84.5% of available credit, ahead of GPT-6 Astra at 83.3%, GPT-6.1 Sol at 82.0% and GPT-5.6 Sol at 58.2%. Across 11 GPT models, OpenAI reported scores from 35.02% to 83.33% at xhigh; the results are company-reported, not an independent comparison with rival systems. The benchmark matters to drug developers because it probes decisions that can guide which compounds scientists test, but strong overall scores do not mean reliable predictions: models earned only about half the credit on compound selection and numerical-property prediction. OpenAI says MedChemBench reflects real medicinal-chemistry workflows, from structure reconstruction to lead optimization.