How big was OpenAI's Navier-Stokes agent swarm?

OpenAI says the resolving group involved on the order of 10,000 concurrent agents, which sent 2.7 million messages and used about 130 billion output tokens, reaching the solution in 88 hours plus 17 hours of Lean formalization.

What triggered OpenAI's Millennium Prize effort?

OpenAI's own post says the effort began September 1 after hearing a rumor later connected to NYU's Tristan Buckmaster and Anthropic's Levent Alpöge, who were working on a related forced Euler blowup result.

Terence Tao Warns AI Is Flattening Open Math Problems

Q: What did Terence Tao warn about AI and open math problems?

Tao wrote that even a rumor of someone working on an open problem can now trigger a massive AI-powered effort to 'flatten' it before the original research reaches its full potential, incentivizing mathematicians to stop sharing directions publicly.

Terence Tao posted yesterday evening that the era of sharing promising research directions may be ending. His words: "even the rumor of someone working on a problem can trigger a massive amount of AI-powered effort to flatten it before the original research project has time to reach its full potential." Read that alongside OpenAI's own Navier–Stokes post and the warning stops being hypothetical — OpenAI's post is the case study, filed in the same 48 hours.

The rumor was the trigger

OpenAI's writeup states it plainly: their Millennium Prize effort "began on September 1st after hearing a rumor which we later realized was related to Levent Alpöge, an Anthropic employee, and Tristan Buckmaster." From a rumor alone they deployed a coordinating-agent system that peaked at roughly 10,000 concurrent agents, burned 2.7 million messages and ~130 billion output tokens on Navier–Stokes alone, and reached the solution in 88 hours — with another 17 for Lean verification. Tao's "massive amount of AI-powered effort," quantified.

Flattened in real time

The mechanics matter. Buckmaster and Alpöge had a different result — a forced Euler blowup, built with retail Claude and Codex. OpenAI's swarm, pointed at every Millennium Prize problem in parallel via problem-statement variants, arrived independently but faster and louder, announced first, and set off a public credit fight. That is Tao's scenario playing out step by step: original research mid-flight, a rumor escapes, a swarm industrializes the space, and the original project gets scooped by its own echo. OpenAI disclaims any prize claim and denies data access — but the structural point stands regardless of who did what.

Why it matters

If the rational response to having a good idea is secrecy, mathematics loses the thing that made the last century of progress possible: open problems, shared early, attacked collectively. Tao says the incentives now point toward hoarding — "reverse centuries of tradition." Watch for the tell: fewer seminar talks that open with "here's a problem nobody's solved." When mathematicians start going quiet, that's the signal. Everything published after this week is the control group.

The rumor was the trigger

Flattened in real time

Why it matters

Related