# OpenAI posts 722 math manuscripts — internal model results, some Lean-checked

> OpenAI's biggest math drop yet: 722 AI-written manuscripts on GitHub, with Lean proofs for many but not all.

- Org: OpenAI
- Date: 2026-10-06
- Category: paper, repo
- Source: https://github.com/openai/math
- Canonical page: https://ai-tldr.dev/releases/openai-math-722-manuscripts/

## In short

OpenAI released 722 math manuscripts from an unreleased internal model. On 7 October it withdrew three Hodge-conjecture papers over a sign error and revised 14 others; 719 remain, about 42% with Lean proofs.

## What it is

The openai/math repository is a public collection of AI-written mathematical manuscripts, first released as 722 papers grouped into 372 related families by field. Every result comes from an unreleased internal OpenAI model that OpenAI tests on open research problems. Topics range from number theory and complexity theory to mathematical physics, including the irrationality exponent of π and a relativistic Vlasov–Maxwell system.

## How it works

Most results came from one fixed procedure: OpenAI posed roughly 4,000 problems to the internal model, and each accepted result used on average three hours of ChatGPT Pro thinking compute. The repo has three parts: preprints/ (PDFs and sources), a lean/ formalization library, and reasoning_traces/ with abridged traces for 10 selected results, such as the symmetric Mahler conjectures and Kaplansky's conjecture.

## Why it matters

The first correction round came within a day. On 7 October 2026 OpenAI withdrew 'Algebraicity of Weil classes on split abelian eightfolds' after a sign error broke a key argument, plus two Hodge-conjecture papers that depended on it, and revised 14 more. The catalogue now lists 719 manuscripts, with 300 of 719 top-line results (about 42%) formalized in Lean, so unformalized results still need human checking.

## Who it's for

mathematicians, theoretical computer scientists, formal-verification researchers

## Try it

git clone https://github.com/openai/math

## Quick facts

- Maker: OpenAI
- Manuscripts: 722, in 372 families
- Problems posed: About 4,000
- Compute per result: ~3 hours of ChatGPT Pro thinking (average)
- Model: Unreleased internal OpenAI model
- Repo license: Apache-2.0

## FAQ

### Are all of OpenAI's 722 math manuscripts formally verified?

No. The openai/math README says not all manuscripts have accompanying Lean formalizations, and that some of the unformalized results could have issues. OpenAI says it will keep adding Lean formalizations to the repository as it obtains them, so the share of machine-checked results should grow over time.

### Has OpenAI withdrawn any of the math manuscripts?

Yes. On October 7, 2026, the openai/math release history recorded that three manuscripts were withdrawn, starting with "Algebraicity of Weil classes on split abelian eightfolds". A sign error in that paper breaks a stabilization-trace cancellation argument that two dependent papers on K3 surfaces also rely on. The withdrawn papers now carry notices explaining the gap.

### Which model produced the OpenAI math results, and can I use it?

The OpenAI math manuscripts were produced by an unreleased internal OpenAI model. It is not available in ChatGPT or the API. The README only says each result used, on average, the equivalent of three hours of ChatGPT Pro thinking compute with that model, which describes effort, not public access.

### How will OpenAI handle mistakes in the math manuscripts?

OpenAI says it will preserve the public release history of the openai/math collection. Corrections and revisions will be recorded as new versions, and earlier released versions will stay accessible, so readers can see what changed after review instead of finding results silently edited or removed.

### How is this different from OpenAI's earlier math announcements?

OpenAI's August release covered ten claimed advances in one long document. The October 6 openai/math release is far larger: 722 manuscripts in 372 families from about 4,000 posed problems, split into preprints, a Lean library and reasoning traces, under an Apache-2.0 license on GitHub.

## Links

- [GitHub Repo](https://github.com/openai/math)
- [Release History](https://github.com/openai/math/blob/main/history.md)
- [TechCrunch](https://techcrunch.com/2026/10/08/openais-math-solutions-arent-meeting-the-fields-standards-yet/)
- [Interesting Engineering](https://interestingengineering.com/ai-robotics/openai-largest-math-release-lean-proofs)
- [Superpower Daily](https://superpowerdaily.com/posts/openai-publishes-722-ai-generated-math-manuscripts-with-verification-still-uneven)
