When 10,000 AI Mathematicians Went After NavierStokes How a nearly two-century-old problem reached an 88-hour agentic assault, 130 billion output tokens, and formal verification in Lean

On 8 September 2026, OpenAI announced a result that may become the first Millennium Prize
Problem solution found by an artificial system. Unlike the familiar pattern of “one model -
one answer,” the research was organized as a massive parallel scientific process: groups of
coordinating agents attacked different versions of the main problem and related problems,
used a cached copy of the Internet and code execution, exchanged intermediate findings within
groups, and then transferred successful ideas across groups through a separate consolidation
loop. For Navier-Stokes, OpenAI reports on the order of 10,000 concurrent agents, 88 hours
of search, 2.7 million messages, and roughly 130 billion output tokens. The analytic proof
was then formalized in Lean in an additional 17 hours with GPT-6 Astra. This article explains,
in accessible language, the problem itself, the nearly century-long human mathematical path
toward the present result, the physical and mathematical idea behind the blowup construction,
the role of Lean, and why the most historically important event may be not only an answer to
Navier-Stokes, but the appearance of a new organizational unit of science - a massive agentic
research collective.