FULL STORY

OpenAI Claims Navier-Stokes Proof with 10,000 Agents

OpenAI says an internal model with ~10,000 collaborating agents proved the Navier-Stokes problem, and reports substantial progress on another millennium problem.

2026-09-09 ~ 2026-09-11 · 3 episodes · 20 posts

Episode 1 · OpenAI's 10,000 Agents Crack the Navier-Stokes Millennium Problem in 88 Hours (2026-09-09, 10 posts)

OpenAI claims that an internal model still in training (reportedly stronger than the commercial GPT-6 Astra, per @r0ck3t23) deployed roughly 10,000 collaborating AI agents to solve two key bounds of the existence and smoothness problem for the Navier-Stokes equations within 88 hours, with frontier-evaluation-level safety monitoring and isolation measures throughout. Ben Goertzel estimates OpenAI spent about $22 million in compute on the effort. This marks the first time an AI system has attacked a top-tier mathematical problem at scale, signaling a leap in AI's capacity for deep scientific research.

Confirmed

  • OpenAI officially stated: an internal model solved the Navier-Stokes problem in 88 hours using about 10,000 collaborating agents, and claims "step-change" gains across numerous benchmarks.
  • Historical context: Leray proved weak solutions exist in 1934, but whether solutions stay smooth has remained open for 90 years; in 2000 the Clay Mathematics Institute listed it among its seven Millennium Prize Problems, each carrying a $1 million bounty. Navier-Stokes governs fluid motion, from airflow over wings and blood flow to weather and oceans.
  • @bengoertzel commented that OpenAI devoted roughly $22 million in compute to the solve.

Not Yet Confirmed

  • Specific capability details of the internal model, its gap versus commercial models, and whether it will ultimately pass full peer review to claim the Clay prize all rest on OpenAI's unilateral claims and await independent verification.

Why It Matters

  • @kacoef notes this news shares the same theoretical intersection as the ZrTe₅ quantum experiment breakthrough, hinting AI may reuse the same mathematical capability across multiple frontier sciences.
  • After having Grok review it, @robleclerc argued "as a mathematical endpoint it's no surprise": Córdoba–Martínez–Zoroa had long established a research program of scale-by-scale blow-up plus smooth forcing, and prior work such as Alpög's paved the way—AI's success is an inevitable step on the shoulders of giants, not creation from nothing.

Episode 2 · Existence Proof Effect: Rumor of Solvability Drove OpenAI's Navier-Stokes Attempt (2026-09-09, 3 posts)

Commentators argue OpenAI only attempted the Navier-Stokes problem after hearing rumors it was solvable, likening it to the four-minute mile effect—once feasibility is known, breakthroughs follow, and AI will accelerate this pattern.

Episode 3 · OpenAI Claims Navier-Stokes Proof, Substantial Progress on Another Millennium Problem (2026-09-10, 7 posts)

According to The New York Times, OpenAI said in a Wednesday evening statement that it has completed a proof of the existence and smoothness problem for the Navier-Stokes equations, and that since finishing it has made substantial progress on another Millennium Prize Problem, "deliberating how to share these results appropriately." This marks the second time in a short period that OpenAI has come close to one of the Clay Mathematics Institute's bounty problems, drawing intense attention from both the mathematics and AI communities.

Confirmed

  • In interviews with The New York Times and in its statement, OpenAI said it has completed the Navier-Stokes proof and made substantial progress on another Millennium Problem, but has not disclosed which one.
  • The Navier-Stokes work is connected to NYU mathematician Tristan Buckmaster and his collaborators.
  • Each of the seven Millennium Problems carries a $1 million prize.

Not Yet Confirmed

  • OpenAI has not revealed which other Millennium Problem it has progressed on.
  • The Navier-Stokes proof itself has not gone through full public peer review, so its mathematical correctness remains to be confirmed.
  • According to reports relayed by @basedjensen, after Buckmaster and collaborators made major progress in August, OpenAI announced the answer in less than a week; outsiders (as @socoolandawesome noted) have raised questions about attribution of the Levent/Buckmaster-related work and academic process, and the specific details of the dispute remain to be verified.

Why It Matters

  • The Millennium Prize Problems are among the hardest problems in mathematics; an AI lab claiming to crack them could reshape both mathematical research and how AI capabilities are evaluated. @DrSingularity accordingly predicts at least 3 Millennium Problems may be solved in succession.
  • If the practice of preempting collaborators' results is confirmed, it will spark debate over academic credit allocation and the conduct norms of AI labs.