In August, OpenAI brought together around 40 mathematicians to discuss what should happen if artificial intelligence outpaces human capabilities in mathematics.
On 21 September, the company said it was putting together an independent advisory group of mathematicians to recommend how such results should be shared. It later consulted the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study on best practices for releasing the work.
At 6pm EDT on 6 October 2026, OpenAI published a GitHub repository presenting 372 families of mathematical results generated by an internal frontier model. The repository includes procedures for revising papers and citing them.
OpenAI says each result either resolves, or makes substantial progress on, a major open question in mathematics or theoretical computer science. Its claims include a solution to the four-dimensional Kakeya conjecture, improvements to some of the world’s most important computer algorithms and progress towards the Riemann hypothesis.
The collection includes formal proofs in Lean, a programming language that can check mathematical proofs by computer. OpenAI said many of the results have already been verified in Lean, which tests a proof’s logic.
An OpenAI spokesperson said almost every result came from a single prompt given to a single AI agent. The spokesperson later told Scientific American that some results may have required multiple attempts.
OpenAI said the average result used computing equivalent to roughly three hours of ChatGPT Pro thinking.
Andrew Sutherland, a mathematician at the Massachusetts Institute of Technology (MIT), said the claims should remain unverified unless researchers can reproduce them using a released model. “Until and unless they release the model and people can replicate their results, I think you should treat any claims about one-shotting problems with a single agent as unverified. We should ask for receipts.”
An unnamed mathematician told WIRED: “There's a perception of mobster behavior” from leading AI companies.
OpenAI’s handling of the release of a solution to the Navier-Stokes problem has already proved controversial. That earlier result was produced through the collective work of a 10,000-strong agentic swarm, using millions of dollars in computing power.
The company said it would fund workshops, conferences and special programmes aimed at understanding major AI-produced results. It also committed to improve its papers through better citations, mathematical exposition and presentation.
OpenAI said it was working towards a responsible release of the model that produced the results.
Comments (0)