Overview
- The GitHub repository released Tuesday, October 6, 2026 contains roughly 722 manuscripts organized into about 372–377 result families and covers problems across number theory, geometry, complexity and mathematical physics.
- Many papers include proofs formalized in the Lean proof assistant, which checks logical steps but does not by itself satisfy institutional processes such as the Clay Mathematics Institute’s prize adjudication or traditional peer review.
- OpenAI provided summary reasoning and aggregate compute statistics—saying an accepted result used about three hours of ChatGPT Pro equivalent on average—but it has not released the internal model, exact prompts or full chains of thought that would allow independent replication.
- Several leading mathematicians have publicly alleged that some AI outputs may have drawn on unpublished human work, a claim the company disputes and that remains unresolved as independent reviewers begin technical checks.
- The company convened an advisory group at the Institute for Advanced Study and independent peer review is under way, but the rapid, large‑scale release is already reshaping debates over attribution, verification and the career effects of machine‑generated mathematics.