OpenAI has released nearly 400 AI-generated mathematical results across more than 700 manuscripts, a flood that researchers say could take years to assess. The scale is remarkable; so is the gap between the claims and the work formally checked so far.
OpenAI Watch analysis
What happened
The collection spans fields from number theory and geometry to theoretical computer science. OpenAI says 300 of 719 manuscripts have been formalised in Lean, a proof assistant that can help verify mathematical arguments. That is about 42 per cent, and formalisation alone does not guarantee that a proof matches the claim made in its paper.
The Verge spoke to more than three dozen mathematicians about the release. Some described early results as highly impressive, including work on major open problems; others said papers were difficult to follow, thinly referenced or not yet verified. Those are researchers’ assessments, not a verdict on the collection as a whole.
Why it matters
AI-generated mathematics is arriving faster than specialists can comfortably check it. That makes the release both a technical milestone and a practical test for how research is reviewed, credited and absorbed. A result that might be important still needs scrutiny; a large pile of results does not come with a matching pile of spare mathematicians.
Our read
This looks like a serious advance wrapped in a very large verification job. Readers should distinguish promising results from confirmed ones, and formal proofs from papers that have not yet had that level of checking. The interesting question is no longer just whether AI can produce mathematics, but how the field can tell which work is sound without making researchers spend years sorting the signal from the noise.
What to watch
- Whether OpenAI adds formalisations for the manuscripts that remain unchecked.
- What independent mathematicians conclude as they examine specific results.
- Whether the release changes expectations around credit, review and publication of AI-generated research.
Discussion spark: Should AI labs release large collections of mathematical results before most have been formally checked, or should verification come first even if that slows discovery?
Sources and evidence
- ‘Pure insanity’: Mathematicians will need years to make sense of OpenAI’s latest drop – The Verge (9 October 2026, 19:09 UTC)
- Source update (9 October 2026, 19:09 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.