OpenAI Says 10,000 AI Agents Solved Navier-Stokes, Mathematicians Dispute Claim
OpenAI announced that an internal model deployed 10,000 autonomous agents to resolve the Navier-Stokes problem, sparking an academic dispute.
OpenAI announced a major milestone on Sept. 8, 2026, stating that an internal artificial intelligence model deployed roughly 10,000 autonomous agents to resolve the Navier-Stokes existence and smoothness problem. The Navier-Stokes equations describe how fluids and gases move, underpinning practical fields such as weather forecasting, aircraft wing design, and ocean current modeling. While engineers use approximations of these equations daily, mathematicians have chased a complete proof for roughly 90 years, ever since Jean Leray showed in 1934 that generalized solutions exist. OpenAI's claimed proof argues that an initially smooth fluid at rest can develop a singularity—a point where velocity becomes infinite—in finite time, resolving specific statements outlined by the Clay Mathematics Institute. The development has immediately triggered an intense academic dispute involving credit, research ethics, and whether artificial intelligence labs are rushing out unverified claims for marketing purposes.
According to Yellow reporting, OpenAI trained its new internal model starting on August 28, 2026. Executive leadership targeted the open Millennium Prize problems on September 1 after hearing rumors that two of them had already fallen. A smaller group of about 100 agents first addressed an unforced Euler question in approximately 50 hours, before the Navier-Stokes group reported an answer on September 5, some 88 hours after launch. Formal machine verification using Lean took an additional 17 hours. BBC coverage noted that the agents exchanged nearly 3 million messages and consumed 130 billion output tokens during the process. Kursiv Media Узбекистан reported that the computation required an estimated median electricity consumption of 10 gigawatt-hours, resulting in a water cooling footprint of roughly 12.5 million litres, alongside compute costs estimated in the millions.
Media additions
The announcement immediately collided with separate research conducted by Tristan Buckmaster, a mathematician at New York University, and Levent Alpöge, an Anthropic researcher working in a personal capacity. As detailed by Yahoo News Singapore and the Hindustan Times, the pair had spent nearly a year working on related fluid equations using OpenAI's Codex tool, releasing Lean-verified proofs for the forced Euler problem just before OpenAI's announcement. Buckmaster stated that he contacted an OpenAI researcher on September 3 after discovering that information about their progress had reached the company. What followed was a contentious exchange regarding potential joint announcements and authorship arrangements.
Buckmaster alleged that OpenAI scientist Sebastien Bubeck pushed to remove Alpöge from a paper because of his employment at a rival lab, and warned that going public could ruin his career. According to The Tech Buzz, Bubeck later described his remarks as a "poor choice of words" intended to warn against unfounded claims. OpenAI maintained that neither its researchers nor its agents saw the pair's work prior to publication, though the company conceded that de-identified product usage data may have helped improve its models.
| Metric / Phase | OpenAI Navier-Stokes Project | Buckmaster & Alpöge Research |
|---|---|---|
| Primary Focus | Navier-Stokes existence and smoothness | Euler and fluid dynamics equations |
| Timeline & Duration | 88 hours of agent search from Sept. 1 to Sept. 5 | Nearly a year of human-led research |
| Verification Method | Lean formal verification system (17 hours) | Lean-verified proofs for forced Euler |
| Resource Scale | 10,000 agents, 130 billion tokens, millions in compute | Individual research funds and Codex tools |
Prominent mathematicians have expressed broader concerns regarding how artificial intelligence companies approach foundational science. Terence Tao warned that using famous open problems as marketing proof points threatens to destabilize the academic ecosystem that produces new techniques and working researchers, comparing the trend to looting an archaeological site. While the Clay Mathematics Institute originally attached a $1 million bounty to each Millennium Prize problem, OpenAI stated it does not plan to claim the award.
What to Watch Next
- Independent peer review and formal verification by external mathematicians to determine whether OpenAI's proof satisfies the strict conditions established by the Clay Mathematics Institute.
- Further academic disclosures regarding the training data windows and whether de-identified user inputs influenced the model's mathematical reasoning capabilities.
- Potential policy discussions within academic institutions regarding credit assignment, AI tool usage, and disclosure frameworks.
- Additional preprints from human researchers studying weakened friction cases that may lead toward the unforced Navier-Stokes problem.
