# OpenAI posted 700 AI-written maths papers. Mathematicians are still reading

URL: https://ai-pulse.aglaganova.com/blog/openai-posted-700-ai-written-maths-papers-mathematicians-are-still-reading
Published: 2026-10-10
Author: Roman Aglaganov, Full-stack developer
Category: Research
Region: Global
Tags: reasoning, science

> An unreleased OpenAI model produced hundreds of manuscripts on open problems, including a claimed proof of the Unique Games Conjecture. Some are checked by computer. Almost none have been understood by a human yet.

![Illustration: a recursive square construction crossed by a red path, with a small human figure looking on](https://ai-pulse.aglaganova.com/api/media/file/math-1.png)

On 6 October, OpenAI opened a public GitHub repository called simply “math”. Inside: more than 700 mathematical manuscripts, written by an unreleased internal model, claiming progress on open research problems across many fields. It is the largest public dump of machine-made mathematics so far, and the maths world has spent the week trying to work out what is in it.

## What OpenAI released

According to the repository, the catalogue now holds 719 manuscripts grouped into 372 “families” of related results, released under the Apache 2.0 licence. OpenAI says it expanded its evaluations to open research problems after its models saturated its existing maths tests. Over the evaluation, the model was given about 4,000 problems, and each result used on average three hours of ChatGPT Pro thinking compute. Almost all results came from the same fixed procedure; OpenAI lists the exceptions.

The repository is upfront about its limits. About 42% of the top-line results come with formal proofs in Lean, a programming language whose checker verifies every logical step. The rest have not been machine-checked, and the README warns that some could have issues.

## The headline claim

Among the families is a claimed proof of Subhash Khot’s Unique Games Conjecture, a central open question in theoretical computer science that, if true, means many optimisation problems are hard to approximate any better than today’s best general methods allow. The computer scientist Scott Aaronson, whose wife Dana Moshkovitz has worked towards it for years, called the release “one of the biggest days in mathematical history”. He also noted that, as far as anyone could tell, no human had yet understood almost any of the proofs.

## Already being corrected

The collection is moving fast. On 7 October, OpenAI withdrew three manuscripts after a sign error broke an argument that two other papers depended on, and revised 14 more with repaired proofs and clearer statements. That is how maths is supposed to work, but it is a reminder that “posted” is not the same as “proved”.

**What to watch:** which results the Lean checker covers, and which ones human experts can follow and confirm in the coming months. Those are the ones that will change textbooks.

## Sources

- [OpenAI: the openai/math repository on GitHub](https://github.com/openai/math)
- [OpenAI: release history of the collection (withdrawals and fixes)](https://github.com/openai/math/blob/main/history.md)
- [Scott Aaronson: The Mathocalypse](https://scottaaronson.blog/?p=10169)