The Judge Settled It Without Naming Anyone

Start with the problem itself. Navier-Stokes asks whether solutions to the equations describing fluid motion in three-dimensional Euclidean space exist, and whether those solutions stay smooth all the way through. CMI unveiled it alongside the other Millennium Prize Problems at a meeting in Paris in 2000 titled The Universality of Mathematical Thought, attaching $1 million to each problem. The statement repeats that these are not arbitrary puzzles akin to fiendish crosswords but fundamental challenges marking the frontier of human knowledge.

Where prize money sits, rules sit with it, and those rules are the real frame of this news. Under the Millennium Prize rules revised on September 26, 2018, a proposed solution must satisfy three conditions. It must be published in a qualifying outlet, at least two years must have passed since that publication, and it must have received general acceptance in the global mathematics community. If any one is missing, the evaluation does not begin.

There is one more thing. CMI states in the rules that it does not accept direct submission of proposed solutions. There is no counter at which to file a claim for the money. The institute also writes that the rules document is a complete statement of the rules and procedures and that it will not offer any further guidance or advice. The judge stands outside the field, and deciding whether the game is over falls first to the mathematics community.

Exactly one sentence in the statement points at AI. After noting that breakthroughs in the surrounding field in recent years had raised hopes the problem might soon be resolved, CMI adds that "The increasing ability of new technologies to accelerate mathematical research has heightened this sense of anticipation." No company, no model, just the word technologies. It is the most careful phrasing available to an institution that administers $1 million.

The name CMI did not say is already public. On September 8, OpenAI posted a piece titled On the Navier–Stokes Millennium Prize Problem on its own site, announcing that an internal system had resolved the problem with an analytical proof and a Lean formalization. The company said it used an internal model significantly more capable than GPT-6 Astra, and that the model's training began on August 28. It released the proof as a PDF alongside a GitHub repository holding the Lean-formalized version.

The passage describing scale stands out. According to OpenAI, the work ran as a multiagent system of roughly 10,000 concurrent agents split into groups that communicated with one another, and the group that produced the Navier-Stokes result ran for about 88 hours, exchanging 2.7 million messages and consuming roughly 130 billion output tokens. These are not numbers in the same unit as one mathematician working on paper.

What the proof says is not the optimistic outcome. OpenAI explained that it showed an initially smooth fluid at rest can develop a singularity in finite time. A vortex grows in a way the company described as one that "spirals inward and gets increasingly elongated, like spaghetti," so that velocity runs to infinity even though energy and applied forces stay finite. The problem closed on the side where smoothness breaks, not the side where smooth solutions always exist.

Another company closed an adjacent problem in the same stretch. OpenAI's post states that Levent Alpöge and Tristan Buckmaster, a professor of mathematics at NYU, independently resolved the forced Euler problem using an internal Anthropic model. METAL reported on the statement Buckmaster gave about the phone call that came the day before the announcement. Models from two companies closed two neighboring problems in the same week.

So the heaviest sentence in the statement is not about money but about credit. CMI writes that "The rules governing the prizes describe the process for evaluating what has been achieved and for assigning credit," and follows it with "The process is deliberately unhurried, but we will provide updates." Reaching for the phrase assigning credit at all signals that the institute knows this case turns on it. METAL reported that a priority dispute broke out immediately after OpenAI's announcement.

General acceptance is the more interesting condition because no machine can supply it. A Lean-formalized proof can be mechanically checked step by step for truth, but what the rules demand is not verification, it is acceptance. Mathematicians have to read the proof, understand it, and be able to restate it in their own language before the community has taken it in. CMI points at exactly that spot when it writes that "We hope to see waves of new human understanding unleashed as the innovations behind this work are analysed and interrogated."

Signals about how the mathematics community will meet this technology are already out. METAL reported that 25 Fields medalists put their names to a declaration drawing a line for AI companies. The breakthroughs in the surrounding field CMI mentions are not abstract either. At the International Congress of Mathematicians in Philadelphia on July 23, 2026, Deng Yu received the Fields Medal for his work in partial differential equations, and Deng also received a 2026 Clay Research Award together with Zaher Hani.

Setting the rules and the statement side by side, as METAL did, shows where the clock is running. The two years count not from the day a paper went online but from publication in a qualifying outlet. What OpenAI released is a PDF on its own site and a GitHub repository, so the publication clock has not started at all. CMI writing twice that it is unhurried is not modesty; it is a description of procedure.

What remains, where the judge held back, is an order of operations. The machine produced a proof, the rules demand an outlet and two years and acceptance, and acceptance only comes into being when people read. The question standing before who receives the $1 million is whether people can digest this proof, and the institute has handed that homework to the whole community. A statement that named no one is not silence. It is notice to read the rules before the judging starts.