What the Navier-Stokes problem is — and why OpenAI won't claim the prize
The Navier-Stokes equations describe how fluids such as water and air move. Whether their solutions in three dimensions always stay smooth, or can suddenly blow up to infinity, is one of the seven Millennium Prize Problems, each carrying a 1 million dollar prize. On September 8, 2026, OpenAI published a 166-page manuscript and a machine-checkable proof in the Lean language showing that an initially smooth fluid can develop a singularity in finite time when a smooth external force is applied. An internal model and about 10,000 agents produced it over 88 hours. OpenAI said it will not claim the prize, and the Clay Mathematics Institute has begun a review
The three lines
- The equations — describe fluid motion; whether 3D solutions always stay smooth is a Millennium Prize Problem worth 1 million dollars
- The claim — on September 8, 2026 OpenAI published a 166-page manuscript and a Lean proof of blow-up under a smooth external force
- The catch — whether a forced solution meets the prize criteria is open; OpenAI will not claim it, and the Clay Institute is reviewing
Key questions
- What are the Navier-Stokes equations
- **Newton's laws of motion applied to fluids.** They are used to model flowing water, spreading smoke and air over an aircraft wing. Claude-Louis Navier and George Stokes completed them in the first half of the 19th century. | Item | Detail | |---|---| | What they describe | how velocity and pressure at each point of a fluid change over time | | Forces included | inertia, pressure, viscosity, external forces such as gravity | | Used for | weather forecasts, aircraft and car design, blood flow, ocean currents | | Open question | whether solutions **always exist and stay smooth in 3D** | **Engineers solve these equations numerically every day.** Nobody had proved that the solutions always make mathematical sense — that velocity never shoots to infinity at some point. That is the Millennium Problem.
- What are the Millennium Prize Problems and how many are solved
- **Seven problems announced by the Clay Mathematics Institute in 2000, each with a 1 million dollar prize.** | Problem | Status | |---|---| | Poincaré conjecture | **solved** — Grigori Perelman, who declined the prize | | Navier-Stokes existence and smoothness | **OpenAI claim, September 2026, under review** | | Riemann hypothesis | open | | P versus NP | open | | Yang-Mills existence and mass gap | open | | Hodge conjecture | open | | Birch and Swinnerton-Dyer conjecture | open | **The Navier-Stokes problem can be settled either way.** Prove solutions always stay smooth (statements A and B), or prove they can break down (C and D). OpenAI claims to have proved **C and D — that breakdown can happen.**
- Why is OpenAI not claiming the prize
- **Because of the external force.** In OpenAI's construction, a **smooth external force** acts on the fluid throughout. Under that condition it shows a singularity forms. | Point | Detail | |---|---| | What OpenAI showed | under a smooth external force, an initially smooth fluid develops a singularity in finite time | | Open question | whether a forced solution satisfies the Clay Institute's formal criteria | | OpenAI's position | **it does not intend to claim the prize** | | Clay Institute | acknowledged the possible resolution and **began expert review** | The Next Web called the decision not to claim **the clearest available signal** of how OpenAI rates its own claim. Quanta reported that formal verification in Lean gives mathematicians confidence the proof is correct. **A correct proof and a prize-qualifying proof are different questions.**
- How did AI produce the proof
- **Not one model thinking for a long time, but thousands working in parallel.** | Item | Figure | |---|---| | Period | September 1–5, 2026, **88 hours** | | Agents | about **10,000** running concurrently | | Messages exchanged | about **2.7 million** | | Tokens used | about **130 billion** | | Extra verification | **17 hours** with GPT-6 Astra | | Released | 166-page manuscript, **Lean formal proof**, GitHub repository | **Lean is a language in which a computer checks a proof line by line**, so no one has to read all 166 pages to find gaps (see 「What a formal proof is」). Critics note the approach requires computing resources only well-funded labs possess. It follows Anthropic's Claude formalising Fermat's Last Theorem in early September.
The equations for flowing water are two centuries old. Whether they always make sense is still unknown. In September 2026 came a claim that AI had answered half of that question.
1. Definition — the equations of fluid motion
The Navier-Stokes equations describe how the velocity and pressure of a fluid — water, air, oil — change over time. They apply Newton's second law (force equals mass times acceleration) to flowing matter.
| Term | Meaning | Example |
|---|---|---|
| Inertia | moving fluid keeps moving | a whirlpool persists |
| Pressure | pushes from high to low | wind |
| Viscosity | internal friction slows motion | honey flows slowly |
| External force | pushes from outside | gravity, a stirring spoon |
Weather forecasts, aircraft wings, car aerodynamics and blood flow are all computed with them. Yet no one has proved mathematically that solutions always exist and stay smooth.
2. What the problem asks
In three dimensions, can a smoothly flowing fluid ever develop a point where velocity becomes infinite?
| Possibility | Meaning | Millennium statements |
|---|---|---|
| never breaks down | solutions stay smooth forever | A and B |
| can break down | a singularity forms in finite time | C and D |
Proving either qualifies. The Clay Mathematics Institute named it one of seven Millennium Prize Problems in 2000, with 1 million dollars on each. Only one — the Poincaré conjecture, solved by Perelman in 2003 — has been resolved.
3. What OpenAI claims
| Item | Detail |
|---|---|
| Announced | September 8, 2026 |
| Result | under a smooth external force, an initially smooth fluid reaches a singularity in finite time (C and D) |
| Released | 166-page manuscript, Lean formal proof, GitHub repository |
| Method | unreleased internal model plus about 10,000 agents, 88 hours |
| Checking | 17 more hours with GPT-6 Astra |
| Prize | will not claim it |
| Clay Institute | review under way |
4. The key caveat — the external force
In OpenAI's construction the fluid is pushed by a smooth external force throughout — water that someone keeps stirring. Under that condition, it blows up.
| Question | Answer |
|---|---|
| Is the proof correct | formal verification in Lean makes that likely (Quanta) |
| Does a forced solution meet the prize criteria | undecided — mathematicians must judge |
| OpenAI's own view | declining the prize signals it is not sure |
The proof is correct and OpenAI solved the Millennium Problem are different sentences. A credit dispute also followed involving Levent Alpöge of Anthropic and Tristan Buckmaster of NYU; OpenAI later acknowledged their concurrent work.
5. Why it matters
| Significance | Explanation |
|---|---|
| Method | a swarm of 10,000 agents, not one model, attacked a famous problem |
| Trust | machine verification in Lean, not only human referees, underpins confidence |
| Cost | 130 billion tokens — possible only for organisations with vast compute |
| Trend | the second major case after Claude formalised Fermat's Last Theorem in early September |
6. Common questions
Will weather forecasts now be wrong? No. Forecasts are numerical approximations; this is a theoretical result about specific mathematical conditions.
Who gets the prize? OpenAI says it will not claim it. That depends on the Clay review and the resolution of the credit dispute.
7. What remains unconfirmed
- The scope of the prior work — unforced problem or forced Euler equations — is reported inconsistently.
- The Clay review timeline is not public.
- Whether the Lean proof formalises every step of the manuscript was not independently checked.
Sources
- Quanta Magazine — AI has solved one of math's $1 million Millennium Prize problems
- The Next Web — OpenAI publishes its Navier-Stokes proof and says it will not claim the Millennium Prize
- CNN Business — OpenAI says it has solved one of math's Millennium Problems
- Axios — OpenAI's historic math solution overshadowed by credit controversy
- Wikipedia — Navier–Stokes existence and smoothness
- Wikipedia — Millennium Prize Problems