OpenAI introduces MedChemBench for medicinal chemistry reasoning
Original title: Can AI help scientists identify promising drug candidates?
Ninety-nine problems from 117 papers grade structure and property prediction; GPT-6.2 Astra scored 84.5%.
OpenAI released MedChemBench, a benchmark that tests whether models connect molecular structures with experimental evidence, explain results, compare compounds, and predict properties. It includes 99 problems, 761 molecular structures, and material from 117 papers and patents. At the xhigh setting, GPT-6.2 Astra scored 84.5%, ahead of GPT-6 Astra at 83.3% and GPT-5.6 Sol at 58.2%.
Unconfirmed: Scores are OpenAI's reported results on its own benchmark.
- Graded areas are structures, reasoning, ranking and selection, and property predictions.
- GPT-6 Astra led at reconstructing structures and explaining experimental evidence.
- Selecting compounds and predicting numerical properties was harder for GPT-6 Astra.
- GPT-6 Astra rebuilt one 18-atom-ring compound correctly in all ten responses.
- Products
- MedChemBench
- Organizations
- OpenAI
From the original article
7 min read at openai.comMedChemBench evaluates how well AI models interpret molecular structures, reason from experimental evidence, and select compounds in early drug discovery.
openai.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.