OpenAI’s MedChemBench Scores GPT-6.2 Astra at 84.5%
The benchmark evaluates 99 drug discovery problems across molecular structures, evidence-based reasoning, compound selection and property predictions.

OpenAI introduced MedChemBench, a benchmark that evaluates how AI models handle early drug discovery tasks, with GPT-6.2 Astra earning the highest score among the models listed at 84.5% of available credit.
The benchmark contains 99 problems drawn from 117 scientific papers and patents and covers 761 molecular structures. It evaluates models on reconstructing molecules from chemical drawings, explaining choices with evidence, ranking compounds against research goals and predicting molecular properties.
GPT-6 Astra scored 83.3%, GPT-6.1 Sol scored 82.0% and GPT-5.6 Sol scored 58.2%. The results varied by task: GPT-6 Astra performed well at reconstructing molecular structures and explaining experimental evidence, while selecting compounds and predicting numerical properties were more challenging.
Some problems require weighing competing properties such as efficacy, absorption and stability. A compound may act on a disease-related protein but be poorly absorbed or break down quickly, creating trade-offs for researchers to assess.