OpenAI publishes Navier–Stokes Millennium claim with Lean formalization

8 Sep 2026: OpenAI published that an internal multi-agent system produced a proof and Lean formalization claiming finite-time singularity for 3D Navier–Stokes — resolving Clay statements “C” and “D” on its reading. Company page is the primary. Mathematical acceptance by Clay / the community is not confirmed.

8 Sep 2026: OpenAI published “On the Navier–Stokes Millennium Prize Problem.” The company says it is sharing a solution to the Navier–Stokes existence and smoothness problem — one of the Clay Mathematics Institute Millennium Prize Problems — produced by an internal OpenAI system, with both a writeup and a Lean formalization. That dated company page is the filing event.

OpenAI’s stated result: the dynamics of the Navier–Stokes equations for fluid motion can develop a singularity in finite time. Specifically, the company says an initially smooth fluid at rest can develop a singularity under a smooth applied force while energy remains finite — which OpenAI says establishes statements “C” and “D” in the official Clay formulation.

OpenAI describes the solution as a vortex that spirals inward and elongates. The technical challenge, on the company’s account, is breakdown through the fluid’s own motion, with terms that become large yet cancel so the external force stays smooth while velocity grows without bound.

OpenAI: the agents used an internal model it says is significantly more capable than GPT‑6 Astra, with training ongoing since about 28 Aug 2026. The company says the effort launched 1 Sep after rumors of Millennium resolutions. The group that produced the Navier–Stokes resolution involved on the order of 10,000 concurrent agents; OpenAI says the agents arrived at the resolution on 5 Sep, about 88 hours after launch. Lean formalization and verification took about 17 additional hours via GPT‑6 Astra. Those are company figures.

OpenAI’s page gives token and message counts: about 2.7 million messages and about 130 billion output tokens for Navier–Stokes; about 4.9 million messages and about 300 billion output tokens across attempted problems. Those are company figures — not independently audited, and not a cost estimate.

OpenAI says it does not intend to claim the Millennium Prize for this result. It frames the release as a snapshot of AI progress and of responsibility / pacing.

Concurrent work, on OpenAI’s account: the company recognizes the priority of Levent Alpöge (Anthropic) and Tristan Buckmaster (NYU) on forced Euler, and says its own Euler result was the unforced variant. OpenAI says that after Lean verification on 6 Sep it offered a concurrent release, and that its researchers and agents did not see Alpöge and Buckmaster’s work through any means before public release. Treat that as OpenAI’s narrative. Priority disputes and “influence” allegations in secondary press stay attributed to those parties — this desk does not adjudicate them.

CONFIRMED for this filing: OpenAI published the claim on its site. Quanta (8 Sep 2026) and BBC corroborate the publication event.

NOT confirmed: Clay Mathematics Institute acceptance, community peer review that the Lean statement matches the intended Millennium problem, Fields or prize eligibility, or that the mathematical community has closed the problem. Quanta notes that Lean checking increases confidence but that further scrutiny remains. This filing does not treat “Millennium solved” as settled mathematics.

A dated OpenAI primary puts a Lean-backed singularity claim on a Clay Millennium problem and an internal model beyond GPT‑6 Astra — a capability-and-epistemics filing. Readers get the company’s numbers and concurrent-work caveats without treating Clay acceptance as done.