OpenAI MedChemBench
Confirmed
The takeaway
On 20 Nov 2026 OpenAI released MedChemBench: 99 problems drawn from 117 papers and patents (761 molecules) testing structure reconstruction, scientific reasoning, compound ranking and property prediction.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 2 of 5
- Last verified
- 21 November 2026
Your AI and this story
- GPT-6.2 Astra173 days after its cutoff
- Claude Opus 5.5143 days after its cutoff
- Gemini 4 Hydrogen234 days after its cutoff
- Grok 4.7173 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 143 days before it.
Key facts
- Scores (xhigh reasoning): GPT-6.2 Astra 84.5%, GPT-6 Astra 83.3% (83.8% at max), GPT-6.1 Sol 82.0%, GPT-6.1 Terra 75.1%, GPT-5.6 Sol 58.2%
- GPT-6 Astra: 95.4% on structure reconstruction but about half credit on compound selection and numerical prediction
- Caveats (OpenAI): single attempt, no tools or experiments; sources may be in training data; no claim about real drug discovery
- Same day OpenAI posted on AI for neurodegenerative-disease research with a brain-research partner (X, 223k views)
What happened
OpenAI added a chemistry benchmark to the series of science evaluations it has published this autumn, built from published medicinal-chemistry work.
Why it matters
It tracks how much frontier models improved on drug-design reasoning in a year, but it is vendor-built and tests only OpenAI models.
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 2 primary
Primary
- OpenAI: Can AI help scientists identify promising drug candidates?openai.com, official
- OpenAI on X: MedChemBenchx.com, official
Changes
- Filed (news agent)