Reports from 8–9 September 2026 describe an OpenAI-linked Navier-Stokes result. One source frames it as rumor and highlights an unresolved private-session allegation; another catalogues tweet claims of ~10,000 agents and a year of multi-agent RL while noting no theorem statement or artifact; a third asserts a simultaneous Lean 4 formal proof for a 166-page paper verified in 17 hours; a fourth treats the proof as impressive but human-directed. Mathematical acceptance and theorem scope remain unresolved in the early record.
The event is not an ordinary capability announcement because the reported package couples a large multi-agent inference run on a Millennium-class problem with a machine-checkable Lean 4 artifact. The primary source treats the mathematical claim as rumor, while a later same-day post asserts that OpenAI announced a proof and posted a Lean 4 formal proof alongside a human-readable paper. The novelty, if borne out, lies in that coupling rather than in another benchmark delta.123
Context
The primary document records the claim as unsettled and redirects attention to whether OpenAI used private Codex sessions belonging to Tristan Buckmaster and Levent Alpöge. The AINews roundup, by contrast, catalogues an OpenAI-linked tweet claim of roughly 10,000 agents, about a year of multi-agent RL, and parallel test-time compute, but explicitly notes that no theorem statement, preprint, proof sketch, formal-verification artifact, or referee commentary appears in its tweet set.12
Assessment
The constraint that moved, according to the formal-methods post, is the cost of producing and checking formal proofs: a 166-page paper is said to have been Lean-verified in 17 hours against a 2005 estimate of one work-week per undergraduate textbook page. The AINews framing moves a second constraint: from pretraining-scale model size toward inference-time coordination of many agents. The multiplier post qualifies the second move by insisting that professional mathematicians supplied the problem understanding and direction.123
Caveats
AI hypothesisWhat would change this assessment is a public theorem statement, a preprint or formal artifact independent of the announcement, referee commentary, and a clear account of human versus model contribution. The documents do not agree on whether the result addresses the full 3D incompressible Navier-Stokes global-regularity problem, a related singularity question, or some narrower statement; until that scope and the authorship record are fixed, the mathematical claim remains unverified even if the systems claim is directionally credible.123
Cross-source check
4 sources
Agreed
A Navier–Stokes-related proof or result was publicized by OpenAI or OpenAI-affiliated accounts in early September 2026.Latent Spacejohndcook.comjohndcook.com
The claim concerns the Navier–Stokes equations and a long-standing mathematical problem.johndcook.comLatent Spacejohndcook.comjohndcook.com
AI systems were involved in the reported work or its checking.johndcook.comLatent Spacejohndcook.com
Early accounts did not report community acceptance or independent external validation of the mathematical result.johndcook.comLatent Space
The public discussion included questions about data provenance, authorship, or disclosure.johndcook.comLatent Space
Disputed
Evidentiary status of the proof claim
johndcook.com
Frames the matter as rumors about a possible solution and focuses on an unresolved private-session allegation; it does not assert that a proof exists or has been verified.
Latent Space
States that the tweet set contains no theorem statement, preprint, proof sketch, formal-verification artifact, benchmark report, or independent referee commentary, and warns that 'solution' is ambiguous.
johndcook.com
Asserts that OpenAI announced a proof and posted a Lean 4 formal proof alongside the conventional proof, with the 166-page paper Lean-verified in 17 hours.
johndcook.com
Calls the recent Navier–Stokes proof impressive but offers no artifact or verification detail.
Scale and method of AI involvement
johndcook.com
Mentions only a possible OAI resolution and possible use of private Codex sessions; gives no agent count or training method.
Latent Space
Reports an OpenAI-linked claim of about 10,000 agents, roughly one year of multi-agent RL, and unstructured parallel test-time compute, while saying the human/model division is unspecified.
johndcook.com
Focuses on Lean 4 formalization and verification cost; does not mention agents, reinforcement learning, or test-time compute.
johndcook.com
Argues AI did not initiate the proof; professional mathematicians directed AI to pursue their ideas at speed.
Scope of the mathematical claim
johndcook.com
Says the rumors concern one of the Millennium Prize problems and technical properties of Navier–Stokes solutions, without confirming a solution.
Latent Space
Says public readers understood the claim as the existence/singularity problem, but the exact theorem statement and proof scope are not supplied and the term 'solution' is ambiguous.
johndcook.com
Says OpenAI settled a long-standing question about the Navier–Stokes equations from fluid dynamics, without explicitly identifying the full Millennium Prize formulation.
johndcook.com
Refers to top-tier mathematicians using AI to settle long-standing conjectures and to the recent Navier–Stokes proof, without a theorem statement.
Attribution and human role
johndcook.com
Raises whether OpenAI used Buckmaster and Alpöge's private Codex sessions and notes that OpenAI had not responded at the time of the post.
Latent Space
Discusses NDA and disclosure norms, saying the exact role of humans versus models is unspecified and that 'multi-agent solved X' muddies standard contribution.
johndcook.com
Asserts that professional mathematicians were needed to direct AI and that AI did not initiate the proof.
johndcook.com
Attributes the work to OpenAI's announcement and does not discuss individual human authorship or private-session allegations.
Reported by one source only
Terence Tao wrote a post on the Navier-Stokes problem the day before the rumor post.johndcook.com
Tristan Buckmaster alleged that OpenAI's Navier-Stokes resolution may have used information from his and Levent Alpöge's private Codex sessions.johndcook.com
Buckmaster asked OpenAI whether they used his private sessions; OpenAI had not responded at the time of the post, though the post notes an update with a statement from OpenAI.johndcook.com
Ethan Knight was the source of the claim that about 10,000 agents, roughly a year of multi-agent RL, and unstructured parallel test-time compute were involved.Latent Space
The '88 hours' detail appeared only in a satirical post in the tweet set and should not be treated as confirmed.Latent Space
The tweet set contained no theorem statement, preprint, proof sketch, formal-verification artifact, benchmark report, or independent referee commentary.Latent Space
OpenAI posted a Lean 4 formal proof simultaneously with the conventional human-readable proof.johndcook.com
The human-readable paper was 166 pages, and Lean verification took 17 hours.johndcook.com
Barendregt and Wiedijk estimated in 2005 that formalizing one page of an undergraduate mathematics textbook takes approximately one work-week.johndcook.com
Extrapolating from that estimate, formalizing the 166-page paper by hand would take about 132,800 person-hours.johndcook.com
The author argues AI did not initiate the proof any more than LaTeX did; professional mathematicians directed AI.johndcook.com
The author reports receiving supposed proofs of open conjectures from naive amateurs using ChatGPT and finding them rubbish.johndcook.com
David Smith, a retired print technician, discovered a single shape that creates an aperiodic tiling of the plane in 2022.johndcook.com
There are rumors that a long-standing math problem, one of the Millennium Prize problems, has been solved.
The problem concerns technical properties of solutions to the Navier-Stokes equations [1], a set of equations that describe the dynamics of fluid flow. Popular accounts of the problem are often oversimplified and misleading.
Some reports will speak of the problem as “solving the Navier-Stokes equations.” The task is not to write down a closed-form solution, which can’t be done, or solve the equations numerically, which has been done for decades. The problem is to prove theoretical properties of solutions which are of little interest in practice.
There has been progress toward settling the Navier-Stokes problem. Terence Tao wrote a post on this yesterday.
What’s also interesting is the intrigue around the possible solution. A post this morning says
If I am reading this correctly, Tristan Buckmaster is alleging OAI has a resolution of Navier-Stokes … which maybe used info from Buckmaster and Levent Alpöge’s private Codex sessions.
Buckmaster asked OpenAI whether they used his private sessions and they have not responded. Update: Statement from OpenAI.
Personal note
This topic connects parts of my career spanning decades. My graduate work was in PDEs and I had some interest in the Navier-Stokes equations. Here are some notes I wrote back in the day.
Now I work more with privacy than with PDEs. The question of whether OpenAI uses private data, contradicting their stated policy, is more relevant to my current work than whether the Navier-Stokes equations have global regular solutions.
Related posts
[1] I never know whether to say equation or equations. You’ll hear both. You could think of Navier-Stokes as one vector-valued PDE or three scalar-valued equations.
The source text above remains the property of its publisher and is reproduced for commentary.
OpenAI used roughly 10,000 agents, 130B tokens, and 88 hours to achieve a Navier–Stokes singularity result, claiming a possible second Millennium Prize solution, pushing the frontier of AI-driven mathematical research.
Focuses on the overlooked aspect of OpenAI's announcement: they released a Lean 4 formal proof alongside the human-readable proof, and discusses the dramatic cost reduction in formal verification.
![[AINews] OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next, roughly 10,000 agents and 130B tokens (>$40M), a contender for second ever Millennium Prize awarded](/media/ca/caca47273bc787d5832afaa9a4ad4d3e01d9eddf570ed1ed905a0506f4ac5880.webp)