Kingy AI
Work with Kingy
AI News

OpenAI Says It’s Close to Another Millennium Prize Problem. Is It Hodge?

September 17, 2026 AI News
Abstract geometric forms and connected AI proof nodes represent reported progress toward another Millennium Prize problem.

OpenAI has not announced a second Millennium Prize solution.

The company has made a narrower claim: after publicizing work on the Navier–Stokes problem, it says it made “substantial progress” on another of the seven famous problems. The Information reports, citing a person with knowledge of the work, that OpenAI employees expect the Hodge Conjecture to be solved relatively soon.

That makes Hodge the leading reported candidate. It does not make Hodge solved. There is no public OpenAI paper, named theorem, second Lean formalization, or Clay Mathematics Institute certification.

The evidence so far

Status What the public record supports
Confirmed OpenAI says it made substantial progress on an unnamed Millennium problem. It has released a claimed Navier–Stokes proof and Lean formalization.
Reported The Information says an unnamed source expects Hodge to be solved soon.
Unconfirmed Claims that Hodge is already solved, that Birch and Swinnerton-Dyer is also near completion, or that an internal model called “Aeon” is responsible.

The accurate headline is therefore: OpenAI says it is close to another Millennium problem, and Hodge is the strongest public lead.

What OpenAI has put on the record

On 8 September, OpenAI published its account of a Navier–Stokes effort carried out by a large internal multi-agent system. The company says the system produced an analytical proof and then a Lean formalization showing that an initially smooth, three-dimensional incompressible flow can develop a finite-time singularity.

OpenAI says the work addresses Clay’s statements C and D, the two breakdown routes in the official problem description. Its account says the Navier–Stokes work involved roughly 10,000 concurrent agents, 2.7 million messages, and around 130 billion output tokens. OpenAI says the resolution took about 88 hours to find and another 17 hours to formalize in Lean.

The company also says it does not intend to claim the Millennium Prize for the result. Clay’s rules explain why: a proposed solution must be published in a qualifying outlet, remain accepted for at least two years, and gain general acceptance in the mathematical community before the institute considers an award.

OpenAI’s second-problem claim is much less detailed. The company has not named the problem or released evidence for it. The New York Times and Washington Post both reported the statement. The Information adds the Hodge lead, but its key source is anonymous.

Why Hodge would be a different test

The seven Millennium Prize Problems span several branches of mathematics. Only the Poincaré Conjecture has previously been resolved.

The Hodge Conjecture belongs to algebraic geometry and topology. In broad terms, it asks how much of the topological structure of a smooth projective algebraic variety can be represented by algebraic cycles, geometric pieces that arise from polynomial equations. The conjecture is known in some special cases but remains open in important higher-dimensional settings, including the dimension-four case highlighted by Clay.

That is a different kind of problem from the Navier–Stokes result OpenAI describes. Navier–Stokes can be attacked through a construction: find an admissible flow whose regularity breaks down. Hodge is a broad structural statement about when topological information must come from algebraic geometry.

If OpenAI has found and formalized a new Hodge proof, that would suggest a capability that transfers across major areas of mathematics. The public record does not yet show that it has.

The Navier–Stokes caveat

The Navier–Stokes headline is easy to overstate.

Clay’s official problem description gives four possible formulations, A through D. A and B ask for global smooth solutions. C and D ask for a breakdown example and permit a smooth externally applied force. OpenAI says its proof takes that route.

If the proof holds, it is a legitimate answer to the problem as Clay formulated it. It does not mean every storm, engine, or turbulent pipe flow becomes singular. It means the equations admit at least one mathematically valid finite-time breakdown construction under the allowed conditions.

The Lean formalization makes the proof machine-checkable, which is valuable. It does not remove the need for expert review. Lean verifies the formal statement and proof term supplied to it. Mathematicians still need to check that the formal statement matches the intended Clay problem, that the assumptions are appropriate, and that the construction is genuinely new.

Charles Fefferman, one of the authors of Clay’s Navier–Stokes statement, told El País that the result should be rewritten in conventional mathematical language and examined by specialists. That is the normal route from a machine-checked artifact to accepted mathematics.

What does “close” mean?

OpenAI has not said which stage its second result has reached. “Close” might mean:

  1. A promising proof strategy.
  2. A new lemma or reduction.
  3. A complete proof in ordinary mathematical prose.
  4. A formal proof in Lean.
  5. A result that has survived independent expert review.
  6. A solution accepted under Clay’s rules.

Those milestones are not interchangeable. An AI lab can be near formalization and still be years away from community acceptance.

This matters especially for Hodge. A system may find a relevant theorem, translate a known special case, or identify a promising reduction without proving the general conjecture. A formal proof can also be correct while proving a weaker or differently scoped statement than a headline implies.

The next disclosure should answer five basic questions:

  • Which Millennium problem is it?
  • What is the exact theorem?
  • Is the complete proof public?
  • Is the formal artifact available?
  • Do independent algebraic geometers agree that it addresses the conjecture itself?

Proof, provenance, and credit

The Navier–Stokes announcement also raised questions about provenance. OpenAI’s work followed public attention to a related forced-Euler blow-up result by NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge. OpenAI says it learned of their result only when it was released, that Buckmaster’s recent prompts could not have influenced its model, and that the proofs differ substantially.

Those claims need released artifacts and independent review. The same standard should apply to a second Millennium claim. Readers need to know what literature the system saw, which human researchers shaped the search, whether the result was independently rediscovered, and who produced the decisive ideas.

A statement signed by 25 Fields Medalists during the Navier–Stokes controversy warned that the goals of AI companies and the goals of the mathematical community can diverge. Mathematics values a correct result, but also understanding, explanation, attribution, and new questions.

The verdict

OpenAI may be close to another Millennium Prize problem. Hodge is the strongest public lead because The Information attributes that expectation to a person with knowledge of the work. The evidence does not establish that Hodge is the target, much less that it has been solved.

The next meaningful update will be a named conjecture, an exact statement, a public proof, a formal artifact, and independent experts willing to stand behind the result. Until then, this is a story about an extraordinary claim and the evidence still missing.

Sources