OpenAI introduces MedChemBench to test AI on medicinal chemistry reasoning
OpenAI introduced MedChemBench, a benchmark testing whether models can connect molecular structures with experimental evidence, explain results, compare compounds and predict properties.
- Size: 99 problems from 117 papers and patents, covering 761 distinct molecular structures. Models usually get the context at a chosen point in a research program.
- Graded parts: molecular structures, scientific reasoning, ranking and selection, and property predictions.
- Results (xhigh reasoning, percent of available credit): GPT-6.2 Astra 84.5%, GPT-6 Astra 83.3%, GPT-6.1 Sol 82.0%, GPT-6.1 Terra 75.1%, GPT-5.6 Sol 58.2%.
- Strengths and weaknesses: Astra earned 95.4% of structure credit but only about half the credit for selecting compounds and predicting numerical properties.
- Limits stated by OpenAI: results do not show whether models can discover new drugs; prior exposure to the published sources cannot be ruled out; each response was a single attempt without tools or feedback.