OpenAI claims to have solved the 200-year-old Navier-Stokes existence and smoothness problem—one of the seven Clay Millennium Prize problems carrying a $1 million bounty. Rather than relying purely on human theoretical insight, the company framed the feat as a milestone for frontier reasoning models tackling fundamental science. According to OpenAI researcher Sébastien Bubeck, the team began training a dedicated mathematical reasoning model on August 28 after catching wind of progress inside rival lab Anthropic.
To secure the formal proof, OpenAI deployed a multi-agent orchestration architecture that scaled from 1,000 parallel agents grinding across 50 hours to a peak swarm of 10,000 agents. As OpenAI Head of Research Mark Chen acknowledged, brute-forcing the search space ran up a compute bill in the millions of dollars. The output culminated in a machine-verified proof formalized in the Lean interactive theorem prover.
The Battle Over Scientific Priority
The announcement triggered immediate friction across the research community over academic ethics and credit allocation. NYU mathematician Tristan Buckmaster asserted that OpenAI rushed its compute sprint only after discovering his collaborative work with Anthropic researcher Levent Alpöge. On Monday, Buckmaster and Alpöge released their own documentation outlining key advances on Navier-Stokes, noting their workflow had utilized Claude and Codex.
Buckmaster alleged that OpenAI attempted to negotiate publication terms, including a proposal where Buckmaster would publish a paper stating an internal OpenAI model cracked the problem while excluding Alpöge as a co-author. Bubeck and OpenAI CEO Sam Altman publicly denied proposing any authorship exclusion, defending their internal pipeline.
"And on Sunday morning we had the final solution, Lean-formalized and everything."
Beyond corporate rivalry and prestige signaling, automated Lean-verified solutions for complex differential equations carry direct commercial stakes. Mastering fluid and non-linear partial differential equations directly impacts computational fluid dynamics, industrial aerodynamics, and high-fidelity physical system simulation. As frontier labs turn foundational mathematics into capital-intensive compute races, the line between genuine scientific discovery and aggressive corporate preemption is growing increasingly thin.