OpenAI says an internal model resolved Navier–Stokes and sets up a mathematics advisory group

The capability claim takes two sentences at the top of an OpenAI post. Training on a new internal model began on August 28, 2026. According to the post, that model has since resolved the Navier–Stokes Millennium Prize problem, along with more than 100 long-standing open problems across most areas of mathematics. OpenAI adds that the speed of this surprised its own mathematicians.

What is claimed, and what is not

The post carries no paper, no proof sketch and no outside review. OpenAI frames the situation as a question of communication rather than of evidence. It describes internal discussions about the best way to inform the mathematical community, and says it is working through how broader deployment of math-related capabilities can be done responsibly.

The model is internal. Nothing in the post offers it to anyone outside the company, gives a date for that, or names a product it will appear in. Anyone choosing a tool for actual work gets nothing usable out of this announcement, and that is worth stating before the headline figure does its work.

The letter that came first

A declaration titled "A Severe Misalignment of AI in Mathematics" appeared on September 11, 2026 with a registered DOI. All 27 signatories are Fields medallists, among them Terence Tao, Peter Scholze, Maryna Viazovska and Cédric Villani.

Their argument is not that the results are wrong. It is that solving famous problems is a tool and a proxy, while the goal is conceptual understanding, and that results announced in a rush leave no time for the writeup, the isolation of new methods and the citation of earlier work by others. The declaration raises attribution and plagiarism questions in the same passage.

The declaration also makes the wider case. Years of training in any field produce understanding and the ability to ask new questions, not only a final answer, and a system that delivers the answer directly cuts that link.

What the advisory group may do

• The group operates independently, may publish advice OpenAI did not ask for, and may comment publicly on the company's effect on mathematics.
• Its members are not paid by OpenAI, and the group chooses its own membership.
• It advises on reviewing and communicating results, on coordinating their release, and on professional standards of mathematical research.
• It will not advise OpenAI on the pace of the company's own internal work in mathematics, which the post states in those terms.

Where the group's reach ends

The final item on that list marks the boundary. The declaration's central complaint concerns pace, meaning results produced faster than the community can absorb them, and pace is the one subject the group is told it will not advise on.

Martin Hairer appears on both lists. He signed the declaration on September 11, 2026 and is among the first members of the advisory group, which is hosted at the Institute for Advanced Study.