OpenAI stated it has made substantial progress on another Millennium Prize Problem. The Clay Mathematics Institute established these seven grand puzzles with a one million dollar bounty in 2000. OpenAI does not want the money and instead uses the problems as a public yardstick for model improvement.
OpenAI has not revealed which problem it tackled. Observers and experts have shared their theories on social media. Discussion has focused on the Hodge Conjecture. This 1950 puzzle addresses counting holes in complex geometric shapes.
William Hodge proposed using algebra instead of geometry to count these holes. No one has proven this shortcut always works. OpenAI has not confirmed this guess.
Rumors also link the work to an unreleased model called Aeon. This rumored model is supposedly built for long time horizon tasks and agentic workflows. OpenAI has not confirmed these rumors either.
On September 8, OpenAI reported that an unreleased internal model produced a Lean verified proof. Lean is software that checks math proofs one logical step at a time. The proof showed that Navier-Stokes equations can blow up into physically impossible infinite speed.
Navier-Stokes equations describe how fluids like air and water move. The equations underpin aircraft design, weather forecasting, and blood flow modeling. They resisted full mathematical explanation for nearly ninety years.
Getting that result required about ten thousand coordinating AI agents working for roughly eighty-eight hours. The announcement faced controversy from New York University mathematician Tristan Buckmaster. Buckmaster released a four-page statement detailing events.
Buckmaster and Anthropic researcher Levent Alpoge quietly chased a related proof for almost a year. They finished by August 22. This occurred before OpenAI researcher Sebastien Bubeck learned about it and raced to publish first.
Buckmaster stated that Bubeck told him to either let OpenAI publish the day after his team or write it up alone. Alpoge works at a rival lab, which prompted Bubeck to allegedly ask why Buckmaster would ruin his career by refusing.
Other labs are also posting major math results. Anthropic stated its Claude model formalized a 358-year-old proof of Fermat's Last Theorem in eleven days. Mathematician Terence Tao warned that this race is burning through math's supply of hard problems faster than new ones appear.
The next known step is whether the Clay Institute, or any independent mathematician outside OpenAI, can confirm even one of these claims. Verification for past Millennium Prize submissions has taken years rather than weeks.
