OpenAI's MedChemBench puts medicinal chemistry reasoning to the test
OpenAI introduced MedChemBench, a benchmark of 99 problems drawn from published medicinal chemistry research, and published scores for several GPT models at the xhigh reasoning setting.
OpenAI has introduced MedChemBench, a benchmark that evaluates AI models on reasoning tasks drawn from medicinal chemistry research. The set contains 99 problems drawn from published research, and it draws on 117 papers and patents covering 761 distinct molecular structures.
What the benchmark measures
The benchmark scores four components: molecular structures, scientific reasoning, ranking and selection, and property predictions. On the structure component, it checks correct atom connections and specified stereochemistry in the molecules a model assembles from the drawings and chemical groups provided.
How the models scored
At the xhigh reasoning setting, GPT-6.2 Astra earned 84.5% of the available credit, followed by GPT-6 Astra at 83.3%, GPT-6.1 Sol at 82.0% and GPT-5.6 Sol at 58.2%. On the structure component at that setting, Astra earned 95.4% of the available credit. At max effort, GPT-6 Astra reached 83.8%.
Why it matters
For teams weighing a model for chemistry work, the benchmark reports how several models handled tasks drawn from published research. OpenAI states the limit of those numbers directly: the results do not establish whether models can discover new drugs or improve laboratory decisions. That makes the scores useful for comparing reasoning on curated problems, and not a substitute for evidence from a laboratory or a discovery program.
Sources
Related reading
3 postsDiscussion
Spotted an error? Send corrections to [email protected] .