Can AI help scientists identify promising drug candidates?

OpenAI
OpenAI introduces MedChemBench, a benchmark testing whether AI models can reason about molecular structures and experimental data to support medicinal chemistry.

Summary

OpenAI introduces MedChemBench, a benchmark that evaluates AI models on reasoning tasks from medicinal chemistry. It tests whether models can connect a molecule's structure with experimental evidence to explain results, compare compounds, and predict properties. The benchmark includes 99 problems drawn from 117 papers and patents, covering 761 distinct molecular structures. It evaluates four components: molecular structure reconstruction, scientific reasoning, ranking and selection, and property prediction. At the xhigh reasoning setting, GPT-6.2 Astra earned 84.5% of 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%. GPT-6 Astra was strongest at reconstructing molecular structures and explaining experimental evidence, but weaker at selecting compounds and predicting numerical properties. The benchmark highlights a central challenge in medicinal chemistry: improving one molecular property can worsen another, so AI must reason about tradeoffs. The results do not establish whether models can discover new drugs, and prior exposure to published sources cannot be ruled out.

(Source:OpenAI)