San Francisco, USA – September 10, 2026
OpenAI’s internal AI system has produced what the company says is a solution to the Navier‑Stokes existence and smoothness problem, one of the seven Millennium Prize Problems that have resisted resolution for roughly 90 years. The claim, detailed in a 165‑page proof and a Lean formalisation, suggests that an initially smooth fluid can develop a finite‑time singularity, a mathematical breakdown in which velocity grows without bound.
Background and Significance
The Navier‑Stokes equations govern fluid motion and are central to weather forecasting, aeronautics, and oceanography. The existence‑and‑smoothness question asks whether solutions remain well‑behaved for all time or can blow up. A resolution would reshape theoretical fluid dynamics and impact practical modeling.
AI has already begun contributing to mathematics, but most efforts focused on assistance—checking proofs or suggesting lemmas. OpenAI’s announcement marks a shift toward AI‑generated original research, echoing recent advances from Anthropic’s Claude system on the Riemann hypothesis.
The AI‑Driven Investigation
OpenAI deployed an internal, unreleased model—more capable than its publicly released GPT‑6 Astra—into a multi‑agent architecture. Roughly 10,000 autonomous agents cooperated over 88 hours, exchanging 2.7 million messages and generating 130 billion output tokens. The workflow combined code execution, literature retrieval, symbolic reasoning, and iterative refinement.
"The scale of coordination required to tackle a Millennium Problem is unprecedented," said Dr Jakub Pachocki, Chief Scientist at OpenAI. "Our agents explored a vast search space, iteratively building and testing conjectures until a viable proof emerged."
The agents produced a human‑readable proof and a Lean formalisation that can be mechanically verified. The Lean code, comprising 13 million lines, was uploaded to a public repository for community scrutiny.
Key Findings
- Singularity Construction: The proof identifies a specific vortex configuration that intensifies under the Navier‑Stokes dynamics, leading to a blow‑up in finite time.
- Formal Verification: Using the Lean proof assistant, the entire argument was checked for logical consistency, with no remaining verification gaps reported by the system.
- Resource Consumption: The effort consumed an estimated $15 million in compute, roughly equivalent to the cost of a mid‑size data‑center cluster running for a month.
Community Response and Next Steps
Mathematicians have reacted with cautious optimism. The Clay Mathematics Institute requires independent verification before awarding the prize. Several experts have pledged to review the Lean formalisation in the coming weeks.
"We will conduct a thorough peer review to assess the validity of the argument," said Prof Teresa Rossi, a fluid‑dynamics specialist at MIT.
OpenAI has not filed for the Millennium Prize, noting that the prize rules stipulate that the claimant must be a human. The company frames the work as a proof of concept for AI‑augmented research.
Implications for AI‑Driven Science
The episode highlights a growing trend: AI systems capable of generating novel hypotheses, testing them, and formalising results. As AI models become more capable, the bottleneck may shift from discovery to verification, underscoring the importance of automated proof assistants and collaborative human‑AI workflows.
Sources
- OpenAI press release, September 10 2026,
- The Journal article, “AI Models Generate Advances in Mathematical Research”, John K. Waters, September 10 2026,
- Clay Mathematics Institute, Millennium Prize Problems,
- Lean proof assistant documentation,
FIRAT Editorial Board
Institutional Research Desk · Foresight Institute of Research and Translation
The collective editorial and research translation board of FIRAT, synthesising peer-reviewed evidence, policy briefs, and division milestones across our seven foundational research pillars.
