OpenAI introduces MedChemBench to assess AI reasoning in medicinal chemistry
- MedChemBench
- OpenAI
- medicinal chemistry
- molecular structure
- AI benchmark

The benchmark draws 99 problems from 117 research papers and patents
It covers 761 molecular structures and four evaluation areas
GPT-6.2 Astra scored 84.5% of available credit at xhigh
OpenAI introduced MedChemBench on November 20, 2026, a benchmark designed to evaluate AI model performance on medicinal chemistry reasoning tasks.
MedChemBench consists of 99 problems derived from 117 published research papers and patents, covering 761 distinct molecular structures. The benchmark evaluates models across four components: reconstructing molecular structures, explaining scientific reasoning with experimental evidence, ranking and selecting compounds, and predicting molecular properties.
At the xhigh reasoning setting, GPT-6.2 Astra achieved 84.5% of available credit, while GPT-6 Astra achieved 83.3%, according to OpenAI. GPT-6 Astra also scored 95.4% in molecular structure reconstruction at the xhigh setting. TokenPost reported that GPT-6.1 Sol scored 82.0% and GPT-5.6 Sol scored 58.2% on the benchmark.
OpenAI stated that building AI capable of connecting a molecule's structure with experimental evidence is a step toward helping scientists develop better medicines. TokenPost noted that some benchmark problems require weighing competing properties such as efficacy, absorption, and stability, where a compound may act on a disease-related protein but be poorly absorbed or break down quickly.




