Can AI help scientists identify promising drug candidates?
Can AI help scientists identify promising drug candidates?
The post introduces MedChemBench, a benchmark of 99 problems drawn from 117 papers and patents that tests whether AI models can connect molecular structures with experimental evidence to explain results, compare compounds, and predict properties. At the xhigh reasoning setting, GPT‑6.2 Astra earned 84.5% of available credit, compared with 83.3% for GPT‑6 Astra, 82.0% for GPT‑6.1 Sol, and 58.2% for GPT‑5.6 Sol, with structure reconstruction a clear strength and compound selection and property prediction proving harder. The authors note the results do not establish whether models can discover new drugs or improve laboratory decisions, citing possible prior exposure to published sources and single attempts without tools or feedback.
Source: OpenAI News · openai.com