OpenAI Publishes Hundreds of AI-Generated Mathematical Results for Public Review

by Ben Voss
OpenAI Publishes Hundreds of AI-Generated Mathematical Results for Public Review

OpenAI Publishes Hundreds of AI-Generated Mathematical Results for Public Review

OpenAI has released more than 300 mathematical research results produced by an internal frontier model, giving researchers access to a large collection of machine-generated conjectures, proofs and supporting material. The release moves AI-assisted science beyond isolated demonstrations and into a workflow where independent experts must verify a substantial body of new work.

OpenAI published the results on October 6, 2026, in the public OpenAI math repository on GitHub. The repository contains manuscripts and supporting proof artifacts from work on open research problems. Its catalogue covers areas including algebra, geometry, number theory, probability and mathematical physics.

What OpenAI released

The repository currently lists 722 manuscripts organized into 372 related families. OpenAI says the collection contains results at different stages of verification, and that not all manuscripts have accompanying formal proofs. The repository also says the model was given approximately 4,000 problems during the evaluation, with an average of about three hours of ChatGPT Pro thinking compute used for each result.

Many results include formalizations written in Lean, an open-source programming language and proof assistant that can be used to check mathematical proofs by computer. OpenAI also published 10 summaries of the model’s reasoning, along with statistics about the attempted problems and procedures for revising and citing the manuscripts.

The collection is not being presented as a completed set of accepted mathematical papers. On October 7, the repository recorded the withdrawal of three manuscripts after OpenAI identified a sign error that invalidated an argument and affected two dependent papers. It also recorded revisions to 14 other manuscripts and added six formalizations. The changes show both the potential value of the release and the need for continuing human review.

Why it matters

Researchers already use software to search large spaces, test conjectures and verify formal arguments. This release raises the scale of that process. If the results withstand review, AI systems could help mathematicians identify useful connections and develop proof strategies more quickly. If errors spread through related manuscripts, the same speed could make unreliable claims harder to track.

OpenAI said it consulted an independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study while developing practices for sharing the work. The company said future releases will continue improving citations, mathematical exposition and presentation, and that it plans to support workshops and other programs focused on understanding AI-generated mathematical results.

How to explore the collection

  1. Open the OpenAI math repository and read its README.
  2. Use the overview and manuscript map to find a result by subject or family.
  3. Check whether the result has a corresponding Lean formalization.
  4. Review the manuscript’s version history and any withdrawal notice before relying on a claim.
  5. Follow the repository’s citation instructions when sharing or checking a result.

The immediate impact is likely to be felt most strongly in research and education rather than in consumer software. The broader development is significant, however: AI systems are increasingly being tested not only as assistants that summarize existing knowledge, but as tools that generate potentially original work for experts to evaluate.

References

OpenAI. (2026, October 6). Sharing AI progress in mathematics. https://openai.com/index/sharing-ai-progress-in-mathematics/

OpenAI. (n.d.). OpenAI math [GitHub repository]. Retrieved October 8, 2026, from https://github.com/openai/math

OpenAI. (n.d.). OpenAI math README. Retrieved October 8, 2026, from https://github.com/openai/math/blob/main/README.md

OpenAI. (2026, October 7). History of revisions and formalizations. GitHub. https://github.com/openai/math/blob/main/history.md

Lean FRO. (n.d.). Lean programming language and proof assistant. Retrieved October 8, 2026, from https://lean-lang.org/

Waldvogel, M. (2026, October 7). OpenAI releases progress on more than 300 math research problems, stunning humans. The Washington Post. https://www.washingtonpost.com/technology/2026/10/07/openai-releases-progress-more-than-300-math-research-problems/