OpenAI publishes a batch of mathematical results from an internal frontier model, on GitHub with Lean formalizations
Start with what OpenAI actually published. On October 6, 2026, OpenAI posted Sharing AI progress in mathematics, opening with a single sentence: it is releasing a broad range of new mathematical results produced by an internal frontier model. OpenAI says that to improve how it shares results with the math community, it has been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, drawing on their advice and public recommendations to shape how these results are released.
How the release is packaged may be the most interesting part. OpenAI says it is publishing the results in a GitHub repository (github.com/openai/math), with protocols for paper revisions and citations, while continuing to explore other community-hosted alternatives that meet the committee's guidelines. For future releases, OpenAI commits to further improving the quality of the papers via citations, mathematical exposition and presentation. Putting math results into a public repository with revision and citation protocols is a very different path from the traditional "write a paper, submit to a journal" route.
The second block is verifiability. As part of the repository, OpenAI says it is sharing formalizations of many of the proofs in Lean. Lean is a programming language that allows mathematical proofs to be checked by a computer — handing the question "is this proof correct?" to a machine rather than relying only on human reading. OpenAI says it will update the repository with more formalizations as it obtains them. For anyone reading these results, this is the hardest layer of evidence currently on offer.
The third block is process transparency, which is unusual in AI research releases. To promote scientific transparency and openness, OpenAI says it is also publishing additional detail about how it obtained the results, in three forms: 10 summaries of the model's reasoning; estimations of compute spent in terms of Pro usage on ChatGPT; and statistics about the number of attempted problems. The single number OpenAI highlights: the average result used the equivalent compute of roughly three hours of ChatGPT Pro thinking.
The fourth block is what happens next. OpenAI says it wants this progress to push the frontier of human knowledge and enable further progress in mathematics, and will therefore fund a series of workshops, conferences and special programs around understanding major results produced by AI, sharing more on that soon. As for the model that produced the results, OpenAI says it is working to responsibly release it, and stresses that this is why it matters to keep evaluating its internal frontier models on mathematics and other sciences.
One caveat: the official post does not state the number of result papers, nor the name of the model. Third-party reports circulating online cite different figures and external mathematicians' scrutiny, but those are not in the official text, so this article does not cite them. Only what the official post itself states is presented here.
🤔 Frequently Asked Questions
Where are these mathematical results published?
Per the post: in a GitHub repository at github.com/openai/math, with protocols for paper revisions and citations. OpenAI says it continues to explore other community-hosted alternatives that meet the committee's guidelines.
What process information did OpenAI disclose?
Three things: 10 summaries of the model's reasoning, estimations of compute spent in terms of Pro usage on ChatGPT, and statistics about the number of attempted problems.
What was the average compute used?
The post states that the average result used the equivalent compute of roughly three hours of ChatGPT Pro thinking.
Why consult an outside body?
OpenAI says that to improve how it shares results with the math community it has been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, drawing on their advice and public recommendations.
Will the model that produced the results be released?
OpenAI says it is working to responsibly release the model that produced these results and stresses why it matters to keep evaluating its internal frontier models on mathematics and other sciences; no timeline or specific form is given.
🛠️ Recommended Tools
- Scientific CalculatorThe practical habit when reading this kind of math release is to check the arithmetic. When you see a constant, a bound or an estimate, re-verify the order of magnitude in a scientific calculator instead of trusting the abstract.
- Percentage CalculatorThe one quantified figure in the official post is an average of roughly three hours of ChatGPT Pro thinking. Turning that into cost or share means getting percentages and ratios right first.
- Word CounterOpenAI says it will improve paper quality through exposition and presentation. When writing your own summaries or reviews, check the length first — clarity and length control are often the same problem.
Summary
Per OpenAI's official post (October 6, 2026), Sharing AI progress in mathematics: OpenAI is releasing a broad range of new mathematical results produced by an internal frontier model. OpenAI says that to improve how it shares results with the math community, it has been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, drawing on their advice and public recommendations. This release publishes the results in a GitHub repository (github.com/openai/math) with protocols for paper revisions and citations, while OpenAI continues to explore community-hosted alternatives meeting the committee's guidelines. As part of the repository, OpenAI is sharing formalizations of many proofs in Lean and will keep updating them. For transparency it published 10 summaries of the model's reasoning, estimations of compute in terms of Pro usage on ChatGPT, and statistics on attempted problems, stating that the average result used the equivalent compute of roughly three hours of ChatGPT Pro thinking. OpenAI says it will fund workshops, conferences and special programs around understanding major AI-produced results, and is working to responsibly release the model that produced them. Note: the official post does not state the number of result papers or the model name; third-party figures circulating online are not part of the official source and are not cited here. Every fact here comes from the official source listed below, with no speculation added.
Sources: " ""OpenAI (official): Sharing AI progress in mathematics"" · ""OpenAI GitHub repository: openai/math"