OpenAI says it has produced a proof resolving a famous question about the Navier–Stokes equations, using thousands of AI agents. The claim could mark a major moment for AI-assisted mathematics, but researchers are still examining whether the proof holds up.
OpenAI Watch analysis
What happened
The Navier–Stokes equations, published in 1845, describe how fluids move and underpin work from weather forecasting to aerodynamics. A related Millennium Prize Problem asks whether solutions always exist and remain smooth under specified conditions. The Clay Mathematics Institute offers a $1 million prize for resolving it.
The Irish Times reports that OpenAI published a 166-page proof in September. It presents a counterexample, known as a “blow-up”, in which fluid speed becomes infinite under the equations’ mathematical conditions. The report says about 10,000 OpenAI language-model agents worked collectively for 88 hours to construct it.
That is a striking account of AI contributing to a difficult mathematical argument. It is not yet a settled result: the article says researchers are scrutinising the paper and accompanying code, and that experts have raised questions about its readability and interpretation.
Why it matters
If the proof survives expert review, it would be a substantial mathematical result and a compelling test of whether AI systems can help generate new knowledge, not merely explain existing work. The reported use of many agents also raises practical questions about how such a result was produced and how other researchers can check it.
For now, the important distinction is between a published proof and a proof accepted by the mathematical community. The equations are used every day, but the prize question concerns what can be shown about them in all the specified cases.
Our read
This is worth following, not celebrating as a solved problem just yet. OpenAI has made a concrete claim and supplied a long proof; the useful next step is independent scrutiny of the argument and code. Mathematics does not award points for page count, however heroic the printer.
What to watch
- Whether independent researchers can verify the proof and its counterexample.
- What the accompanying code contributes to checking the argument.
- Whether the mathematical community reaches agreement on what the result establishes.
Discussion spark: What should count as convincing evidence that AI has helped solve a mathematical problem: a published proof, independent verification, or both?
Sources and evidence
- Source update (19 October 2026, 05:00 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.