OpenAI Says Internal Model Has Solved Over 100 Open Math Problems

An internal OpenAI model has resolved more than 100 long-standing open problems across most areas of mathematics, the company announced in a post introducing a new mathematics advisory group. The model, which OpenAI began training on August 28, also resolved the Navier–Stokes Millennium Prize problem, according to the company's earlier announcement.

The scale of the claim is unusual. The Navier–Stokes problem is one of seven Millennium Prize Problems established by the Clay Mathematics Institute, each carrying a $1 million award, and had stood unsolved since 2000. OpenAI now says the same internal model has gone far beyond that single result. "The pace of its progress in mathematics has surprised the mathematicians within OpenAI," the company wrote. That surprise has triggered internal discussions, OpenAI said, "on the best way to inform the community of the rapid progress to prepare and adapt the field."

The announcement arrives amid growing friction between AI developers and the mathematical community. In a recent open letter titled "A Severe Misalignment of AI in Mathematics," published at mathandai.org, mathematicians raised concerns about the negative externalities of using the solving of open problems as a benchmark for new AI systems. Their argument, as OpenAI summarizes it, is that racing to close famous open questions removes problems that structure careers, graduate training and the field's research agenda — costs that fall on mathematicians rather than on the companies running the benchmarks.

OpenAI's response is an advisory group of nine prominent mathematicians: François Charles (ENS-PSL), Camillo De Lellis (IAS, GSSI), Timothy Gowers (Collège de France, Cambridge), Martin Hairer (EPFL, Imperial College London), Nikhil Srivastava (Berkeley, Simons Institute), Ulrike Tillmann (Oxford, INI), Ravi Vakil (Stanford), Edward Witten (IAS) and Melanie Matchett Wood (Harvard). The roster includes two Fields Medalists — Gowers and Hairer — and Witten, the only physicist to have won the Fields Medal. Vakil and Srivastava are known for work in algebraic geometry and theoretical computer science respectively; Tillmann and Wood work at the interface of topology, number theory and applied mathematics.

OpenAI says the group will serve as "a bridge to the mathematical community and broader public, giving mathematicians a voice in how we move forward." Its mandate covers the review and communication of emerging results: members will help OpenAI assess the significance of new solutions, advise on how to coordinate their dissemination, and advise on academic and professional standards of mathematical research. The group will also advise on how OpenAI's tools can support mathematical research and learning. "We want to put capable tools in mathematicians' hands so they can pursue the questions they know best and develop new ideas," the company wrote.

Structurally, OpenAI has built in several independence guarantees. The group will operate independently from the company. It will have the freedom to offer advice OpenAI has not requested, to comment on OpenAI's impact on mathematics, and to make its advice public. Members will not be paid by OpenAI, and the group can change its own membership as it sees fit. "Its value depends on its members being able to exercise their own judgement and challenge ours," the post states.

One carve-out stands out. "Importantly, the group will not be responsible for advising us on how to pace our internal progress on mathematics," OpenAI writes. The advisory group shapes how results are communicated and deployed, not how fast the underlying research moves. That distinction goes to the heart of the open letter's complaint: the letter's authors object to the benchmarking race itself, not merely to how its outputs are announced. An advisory body with no say over pacing cannot slow the rate at which open problems fall.

The stakes extend beyond mathematics. OpenAI itself frames the matter that way: "Mathematics is a fundamental science, and novel discoveries may result in a wide range of applications. That makes the promise of these capabilities substantial, and their responsible development and deployment important beyond mathematics itself." The company says it is "working through how broader deployment of math-related AI capabilities can be done responsibly," but offers no specifics on timing or form of any public release.

For the research community, the announcement raises verification questions the post does not address. Claims of over 100 resolved open problems currently rest on OpenAI's internal assessments, and the post does not name the problems, describe the proofs, or detail any peer-review timeline. The advisory group's role in reviewing "emerging results" suggests some external assessment process is intended, but no mechanism is specified. How the group's first public assessments arrive — and whether they precede or follow formal publication of the claimed results — will test whether the new structure functions as more than communication management.

OpenAI casts the group as a beginning rather than an answer. "Working with this group is a first step," the post reads. "There are difficult questions ahead about how AI can support mathematical understanding and how the benefits of these capabilities can reach the wider community. We want mathematicians to be at the center of shaping the answers." With a model that OpenAI says is solving open problems at a pace that surprised its own mathematicians, the advisory group's first months will indicate whether independent mathematicians can meaningfully shape the deployment of capabilities they do not control.