aifollow.news
← TrendingLatest development Nov 21 05:17
STORY IN FOCUS

OpenAI releases MedChemBench to assess AI reasoning in medicinal chemistry

MedChemBench contains 99 questions drawn from papers and patents, testing models on molecular structures, experimental evidence, compound selection and property prediction. OpenAI's results show stronger performance on molecular structures but clear weaknesses in compound selection and numerical property prediction. The company says scores on historical questions cannot establish whether models can discover new drugs.

1 developments3 reports2 sources

Event developments

Coverage of the same occurrence shares a node; subsequent developments have their own nodes.

OpenAI publishes medicinal chemistry benchmark MedChemBench

MedChemBench draws problems from 117 papers and patents and covers 761 graded molecular structures. Each prompt stops at a real research decision point, provides the biological target, compounds made so far and selected assay results, and withholds results reported later. The source says GPT-6.2 Astra scores about 84% overall but gets the consequential decision right about half the time.

View source reports · 2 sources · 3 reports
All reports · 3

By original publication time, with each report's bookmarks and feedback preserved.

IT之家 科技新闻 Machine translation

OpenAI releases MedChemBench to assess AI reasoning in medicinal chemistry

MedChemBench contains 99 questions drawn from papers and patents, testing models on molecular structures, experimental evidence, compound selection and property prediction. OpenAI's results show stronger performance on molecular structures but clear weaknesses in compound selection and numerical property prediction. The company says scores on historical questions cannot establish whether models can discover new drugs.

Rohan Paul (@rohanpaul_ai) Machine translation

OpenAI publishes MedChemBench to test medicinal chemistry decisions

MedChemBench draws problems from 117 papers and patents and covers 761 graded molecular structures. Each prompt stops at a real decision point, provides the biological target, compounds made so far and selected assay results, and withholds results researchers reported later. The source says the best-performing model, GPT-6.2 Astra, scores about 84% overall but is right about half the time on the key decision.

Rohan Paul (@rohanpaul_ai) Machine translation

OpenAI publishes medicinal chemistry benchmark MedChemBench

MedChemBench draws problems from 117 papers and patents and covers 761 graded molecular structures. Each prompt stops at a real research decision point, provides the biological target, compounds made so far and selected assay results, and withholds results reported later. The source says GPT-6.2 Astra scores about 84% overall but gets the consequential decision right about half the time.