OpenAI Says AI Solved 90-Year-Old Navier-Stokes Problem

Published: 09 September 2026. The English Chronicle Desk. The English Chronicle Online

OpenAI says one of its most advanced internal artificial intelligence systems has produced a solution to the notoriously difficult Navier–Stokes existence and smoothness problem, completing the work in about 88 hours with the help of roughly 10,000 AI agents working concurrently.

If independently validated, the development would represent a major moment for AI-assisted mathematical research. The Navier–Stokes problem is one of the seven Millennium Prize Problems identified by the Clay Mathematics Institute, a collection of mathematical challenges for which the institute established a $1m prize for each accepted solution.

OpenAI, however, is not claiming the prize at this stage. The company says its objective is to demonstrate the progress of its AI systems rather than immediately seek formal recognition as a solution to the Millennium Prize problem.

The announcement has also been accompanied by a dispute over research priority involving mathematicians Tristan Buckmaster of New York University and Levent Alpöge of Anthropic. Their work on related problems and their use of OpenAI’s Codex have raised questions about whether OpenAI’s researchers or systems could have benefited indirectly from unpublished research. OpenAI denies accessing their work and says its proof differs substantially from the researchers’ results.

A Problem That Has Defied Mathematicians for Decades

The Navier–Stokes equations are fundamental to the mathematical description of fluids such as water and air. They are used to model how fluids move, including phenomena associated with turbulent flow.

The underlying Millennium Prize challenge asks whether sufficiently smooth solutions to the three-dimensional Navier–Stokes equations always exist and remain smooth, or whether the equations can develop a mathematical singularity in finite time. The Clay Mathematics Institute describes the problem as one of the major unresolved questions surrounding the equations that govern fluid motion.

That question has implications well beyond pure mathematics. A deeper understanding of the equations could improve the theoretical understanding of turbulence and fluid behaviour, phenomena relevant to areas ranging from aerodynamics and engineering to weather and physical science.

The problem has resisted generations of mathematical work. Its difficulty is not simply a matter of performing complicated calculations. A successful solution requires a rigorous proof that satisfies the precise conditions set out by the Clay Mathematics Institute.

Thousands of AI Agents Working Together

OpenAI says the breakthrough came after researchers began testing a new internal model that showed unusually strong mathematical capabilities.

Rather than asking a single AI system to solve the problem, researchers organised a large-scale network of AI agents. The agents could pursue different approaches, exchange intermediate findings and build on promising results generated elsewhere in the system.

The group working on the Navier–Stokes problem involved about 10,000 concurrent agents. According to OpenAI, they reached their resolution on 5 September, approximately 88 hours after the agents were launched.

The scale of the experiment is striking. OpenAI says that the agents working specifically on Navier–Stokes exchanged approximately 2.7 million messages and generated around 130 billion output tokens. Across all of the mathematical problems included in the wider effort, the company says its systems exchanged about 4.9 million messages and produced roughly 300 billion output tokens.

OpenAI has estimated that the computational effort required millions of dollars in resources.

The process therefore represents something different from a conventional chatbot answering a difficult question. It resembles a large computational research operation in which thousands of specialised AI instances investigate different possibilities simultaneously and communicate their findings.

From AI Discovery to Formal Verification

OpenAI’s announcement also highlights a second stage of the process: formalisation and verification.

After the agents produced their mathematical result, another internal AI system, GPT-6 Astra, was used to formalise the work in Lean, a computer-assisted theorem-proving system. OpenAI says this verification stage took an additional 17 hours.

That distinction matters because generating an apparently convincing mathematical argument is not the same as producing a proof accepted by mathematicians.

A proof can contain a subtle logical error, rely on an unstated assumption or fail to establish exactly the proposition that needs to be proved. Formal verification can provide an additional layer of checking by translating mathematical reasoning into a form that a theorem prover can examine step by step.

Even so, formal verification does not automatically settle every question surrounding the significance of a result. Mathematicians must still determine whether the formalised statement corresponds precisely to the original Millennium Prize problem and whether all of its assumptions and constructions meet the required conditions.

What OpenAI Says It Has Established

OpenAI says its result demonstrates a finite-time breakdown, or singularity, in the three-dimensional Navier–Stokes setting under the conditions described in its work.

The Clay Mathematics Institute’s official formulation permits several alternative routes to resolving the challenge, including proving global existence and smoothness or demonstrating an appropriate breakdown.

This is an important point because the phrase “solved Navier–Stokes” can make the achievement sound broader than it is. The mathematical problem is a precisely defined question, and satisfying one of its accepted formulations is what matters for the Millennium Prize.

OpenAI has itself avoided claiming the $1m award. The company says the purpose of releasing the work is to report substantial progress in the capabilities of its AI models.

The result therefore remains a claim of a mathematical breakthrough rather than a formally awarded Millennium Prize solution.

A Dispute Over Research Priority

The announcement has quickly become entangled in a separate controversy over how the AI system arrived at its result.

Buckmaster and Alpöge had been working on closely related fluid-dynamics research using AI tools, including OpenAI’s Codex. Buckmaster subsequently said information about their progress had reached OpenAI while they were developing their own work. He questioned the timing of OpenAI’s effort and publicly raised concerns about how the company came to pursue a related direction.

The controversy has raised an unusual question for modern scientific research: where does intellectual priority lie when humans and AI systems are working together across interconnected digital platforms?

The issue is particularly sensitive because AI coding and research systems may process large quantities of information supplied by users. Researchers increasingly use commercial AI tools to explore difficult mathematical and scientific questions, creating uncertainty over the boundary between private research and the development of future AI models.

OpenAI has denied accessing Buckmaster and Alpöge’s work through improper means. The company says it had not seen their work before it was publicly released and maintains that its Navier–Stokes proof differs significantly from theirs. It has nevertheless acknowledged that, while unlikely, de-identified information derived from users’ interactions with its products could have contributed indirectly to model improvement.

Why Independent Review Matters

The most significant next step will be scrutiny by mathematicians outside OpenAI.

Claims involving major mathematical conjectures require exceptional standards of verification because the consequences of an error can be substantial. A proof that appears persuasive to an AI system must survive examination by specialists who understand the underlying mathematics and the precise requirements of the original problem.

The Clay Mathematics Institute remains the organisation associated with the Millennium Prize Problems, including Navier–Stokes. Its official description continues to identify the existence and uniqueness of sufficiently smooth solutions as an unresolved fundamental question.

That means the wider mathematical community will ultimately determine how significant OpenAI’s work is.

The announcement nevertheless demonstrates how rapidly AI-assisted research is changing. Only a few years ago, the idea of thousands of AI agents collaborating on a problem that had challenged leading mathematicians for generations would have seemed largely theoretical. OpenAI’s experiment suggests that increasingly capable AI systems can now coordinate complex mathematical searches at a scale that would be difficult for a conventional research team to reproduce.

A New Model for Mathematical Discovery

The episode may prove important even if OpenAI’s proof requires revisions or fails to satisfy every requirement of the Millennium Prize problem.

AI systems are increasingly being used not merely to calculate answers but to generate conjectures, explore mathematical structures, write computer code and test possible arguments. The combination of autonomous agents, communication between models and formal theorem-proving systems could create a new model for scientific discovery.

At the same time, the controversy surrounding Buckmaster and Alpöge demonstrates that technological capability alone does not resolve questions of authorship, attribution and research ethics.

If AI becomes capable of producing important mathematical results in hours, researchers and institutions will need clearer rules governing data access, confidentiality, intellectual property and scientific credit.

A Landmark Claim Still Awaiting Judgment

OpenAI’s reported 88-hour result is therefore both a technological milestone and an unresolved scientific claim.

The company says thousands of AI agents produced a solution to one of mathematics’ most famous unsolved problems and that a separate AI system subsequently formalised the work. The scale of the computational effort suggests that AI-assisted mathematical research may be entering a new phase.

But the Navier–Stokes problem will not be considered definitively solved simply because an AI system says it has solved it. Independent mathematicians must examine the argument, determine whether it satisfies the exact requirements of the Millennium Prize problem and assess whether the formalised proof establishes the claimed result.

For now, OpenAI has presented its work as evidence of rapidly improving AI reasoning rather than a claim to the $1m prize.

Whatever the final mathematical verdict, the episode has already opened a much larger debate about the future of discovery. The central question is no longer simply whether machines can assist mathematicians, but whether increasingly autonomous AI systems can become genuine engines of mathematical research — and how the scientific community should recognise, verify and govern what they produce.

Checkout our latest news>>