AI news: OpenAI publishes new math results, and the model is not public
On 6 October 2026 OpenAI published a batch of math results from an internal model. Some proofs are being checked in Lean. That model is not available to the public.
On 6 October 2026, OpenAI published a new batch of mathematics results. They came from an internal AI model, not from the model you use in ChatGPT today.
The scale is large, and it is easy to misread. Here is what OpenAI actually said, what is still unsettled, and what it means for everyday users.
What OpenAI announced
In a post titled Sharing AI progress in mathematics, OpenAI said it is releasing a broad range of new mathematical results from an internal frontier model.
Facts from that post:
- The results are published in a GitHub repository, with protocols for revisions and citations.
- Many proofs are being formalized in Lean, a programming language that lets a computer check a proof's logic. OpenAI said it will add more Lean formalizations as they are ready.
- The repository also includes 10 summaries of the model's reasoning, estimates of compute spent, and statistics on how many problems were attempted.
- OpenAI said the average result used compute roughly equal to three hours of ChatGPT Pro thinking.
- The model that produced the results has not been released. OpenAI said it is working on a responsible release.
OpenAI also said this release followed advice from the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study.
Numbers reported elsewhere
The Verge reported that the release contains 722 manuscripts covering 372 result families. Scientific American reported that mathematicians will need months to judge whether the proofs contain new ideas or mostly combine techniques that already exist.
Those figures and readings come from news reports, not from the sentences on OpenAI's announcement page. Treat them as reports still under review.
What this means for you
If you use AI to write, make content, or run daily work:
- The ChatGPT you open today is not this model. Be wary of anyone selling "access to OpenAI's math model" through an unofficial link.
- This is a science update, not a new button in the app. You do not need to change any account setting.
- A proof that passes Lean is strong on logic. Lean does not tell you whether the question matters for your work.
Want tools you can actually try now? See the list on the AI tools page.
Sources
- Sharing AI progress in mathematics (OpenAI)
- OpenAI drops another batch of mathematical breakthroughs (The Verge)
- OpenAI unleashes hundreds more math results upon a field already in shock (Scientific American)