OpenAI Claims Unreleased Model Solved Navier-Stokes Problem, Faces Credit Dispute

OpenAI says an internal AI model — described only as "significantly more capable than GPT-6 Astra" — produced a proof that three-dimensional Navier-Stokes equations can develop a singularity in finite time, a partial solution to one of mathematics' seven Millennium Prize problems. The claim, announced Tuesday, is already contested by outside mathematicians who say OpenAI may have benefited from their unpublished work.

What OpenAI says happened

According to OpenAI, the effort began September 1 after researchers heard rumors that two Millennium Prize problems had recently been solved. The company says it turned its newest, unreleased internal model loose on the remaining problems, deploying roughly 10,000 concurrent agents that reached the Navier-Stokes result in approximately 88 hours.

OpenAI executives told reporters on a call that the effort consumed enormous computing resources, putting the cost in the "millions of dollars." The company says it does not intend to claim the $1 million Millennium Prize associated with the problem, instead presenting the result as evidence of how quickly its frontier models are improving.

No pricing, context window, or public release details have been disclosed for the model used — OpenAI has not made it available via API, and it remains unclear whether or when it will be.

The credit dispute

The announcement has been overshadowed by questions from NYU mathematician Tristan Buckmaster, who — alongside Anthropic researcher Levent Alpöge — had been working on closely related fluid-dynamics research. Buckmaster has publicly asked whether OpenAI's research direction was influenced by rumors or fragments of his and Alpöge's unpublished work, and has raised concerns about whether private Codex material could have played a role.

OpenAI chief research officer Mark Chen pushed back directly on the call with reporters. "No people or AI systems searched through user data to solve this problem," Chen said, adding: "I'm a little disappointed with the allegations." OpenAI maintains that neither its researchers nor its agents saw any of Buckmaster and Alpöge's work before the pair released it publicly.

As of this writing, neither OpenAI nor Buckmaster has released technical details sufficient for outside verification of either the proof or the provenance dispute. The story is developing.

What this means

This is fundamentally a trust story, not yet a verified mathematical breakthrough. OpenAI has not published the proof for peer review, has not named the model, and has not disclosed how or whether it will be released — meaning the claim currently rests entirely on the company's own account. The credit dispute raises a harder question for the field: if frontier labs can rapidly deploy massive agent swarms against open problems, researchers doing slower, incremental work risk having their unpublished efforts overtaken or exposed through incidental data exposure, even unintentionally. Until OpenAI publishes verifiable details — and until Buckmaster's concerns are independently examined — this should be treated as an unconfirmed claim with real institutional-trust stakes, not a settled result.