OpenAI launches MedChemBench to test AI drug chemistry skills
OpenAI has released MedChemBench, a new benchmark that tests how well AI models reason in drug chemistry. It covers molecular structures, reading experimental results, comparing compounds and predicting properties. The benchmark has 99 questions drawn from 117 papers and patents, covering 761 molecular structures.
The test has four parts: molecular structure, scientific explanation, ranking and selection, and property prediction. At the xhigh reasoning level, GPT-6.2 Astra scored 84.5% usable score, GPT-6 Astra 83.3%, GPT-6.1 Sol 82.0%, and GPT-5.6 Sol 58.2%. Astra reached 95.4% on molecular structure, but got only about half marks on compound selection and numerical property prediction.
OpenAI says the models are stronger at rebuilding molecules and explaining evidence, but still have room to improve in compound screening and property prediction. The results cannot show whether a model can find new drugs or improve lab decisions. Each answer used no tools or feedback, and the team cannot rule out that models had seen published material.
Comments
Loading comments…