OpenAI’s GPT-6.2 Astra earned the top score among listed systems on MedChemBench, highlighting how AI models handle evidence, molecular design and property analysis in early drug research.
Astra received 84.5% of available credit on the benchmark. GPT-6 Astra scored 83.3%, while GPT-6.1 Sol reached 82.0% and GPT-5.6 Sol recorded 58.2%.
The evaluation draws on 99 drug-discovery cases sourced from 117 scientific papers and patents and involving 761 molecular structures. It tests how systems reconstruct molecules from diagrams, explain answers using evidence, rank compounds for specific research goals and estimate molecular properties.
Results varied across those tasks. GPT-6 Astra was strongest at rebuilding molecular structures and explaining experimental evidence, while compound selection and numerical property estimates were weaker areas.
Some cases require researchers to weigh efficacy against absorption and stability. A compound may reach a disease target but absorb poorly or break down rapidly, forcing decisions among competing chemical priorities.
OpenAI released GPT-6.2 Astra alongside GPT-6.2 Sol on Nov. 17. EconoTimes previously covered the models’ launch.


