OpenAI Announces Solution to the Navier–Stokes Millennium Prize Problem

TL;DR

OpenAI announced that an internal AI system produced a rigorous analytical proof—and a Lean formalization—showing that smooth three‑dimensional incompressible Navier–Stokes flows can develop a finite‑time singularity, thereby resolving the Navier–Stokes existence and smoothness Millennium Prize problem.

The Problem and Its Significance

The Navier–Stokes equations model fluid motion using Newton’s second law applied to a continuous medium. For three‑dimensional incompressible fluids with constant density, the open question for 90 years has been whether smooth initial data can evolve into a singularity—i.e., velocities that blow up in finite time despite viscous damping. A positive answer would demonstrate a fundamental breakdown of the continuum approximation and would be the first resolution of a Clay Mathematics Institute Millennium Prize problem.

The Announced Result

OpenAI’s system proved that an initially smooth fluid at rest, subjected to a smooth external force, can develop a singularity in finite time while its total kinetic energy remains finite. The proof establishes statements “C” and “D” in the official Millennium Prize formulation, confirming that:

  • The Navier–Stokes dynamics can produce unbounded velocity in finite time.
  • The blow‑up occurs via a vortex that spirals inward and stretches axially, resembling a tightening spaghetti strand.
  • All terms in the equations (acceleration, pressure gradient, momentum transfer, viscosity) become large yet cancel precisely, preserving a smooth external force.

A full write‑up and a machine‑checked formalization in the Lean theorem prover accompany the announcement.

How the Proof Was Discovered

Model Training and Deployment

  • Since 28 August 2026, OpenAI trained a new internal model that outperformed previous benchmarks in mathematics, surpassing the capabilities of GPT‑6 Astra.
  • The model remained in ongoing training, with performance improving throughout the effort.

Multi‑Agent Evaluation Framework

  • On 1 September 2026, rumors of solved Millennium problems prompted OpenAI to launch a coordinated multi‑agent evaluation of all open problems.
  • Agents were organized into groups with internal communication, each group receiving different variants of problem statements (e.g., Navier–Stokes versions “A/B” for a proof, “C/D” for a disproof).
  • The Navier–Stokes effort involved roughly 10 000 concurrent agents.
  • Agents had tool access: cached internet reads, code execution, and later an upgraded model version.
  • After initial successes on the Euler regularity problem (unforced blow‑up), resources were re‑allocated to Navier–Stokes.

Workflow and Resources

  • Agents exchanged 2.7 million messages and generated ≈130 billion output tokens while solving Navier–Stokes.
  • The resolution emerged on 5 September 2026, about 88 hours after the first agents were launched.
  • Formal verification in Lean required an additional 17 hours using GPT‑6 Astra.
  • Across all attempted problems, the system produced 4.9 million messages and ≈300 billion tokens.

Concurrent Work and Attribution

OpenAI learned of a parallel effort by Levent Alpöge (Anthropic) and Tristan Buckmaster (NYU) after completing its own proof. The external team had resolved the forced Euler blow‑up problem, not Navier–Stokes. OpenAI reached out to offer a joint announcement and shared prompt logs, confirming that the two results are distinct.

Implications for Mathematics and AI

  • Mathematical impact – The proof settles a central question in fluid dynamics, confirming that the Navier–Stokes equations can cease to be well‑posed in finite time. This reshapes theoretical work on turbulence, numerical simulation guarantees, and the foundations of continuum mechanics.
  • AI research impact – Demonstrates that large‑scale, multi‑agent AI systems can generate novel, peer‑review‑level mathematics and produce machine‑checked formal proofs. The scale of communication (millions of messages, hundreds of billions of tokens) suggests a new paradigm for automated discovery.
  • Responsibility and future direction – OpenAI emphasizes that it does not claim the US $1 million Millennium Prize and stresses ongoing work to make such systems steerable, accountable, and aligned with human values. The organization plans to study the model’s reasoning processes to guide future capability pacing.

References

Sources