Artificial intelligence may have just cleared a symbolic hurdle. OpenAI says it has made substantial progress on one of the most formidable unsolved problems in mathematics, one that has resisted a complete solution for nearly nine decades.
The company reports that a new internal model, coordinating roughly 10,000 intelligent agents working simultaneously, tackled the existence and smoothness problem for the Navier-Stokes equations. The puzzle is one of the seven Millennium Prize Problems listed by the Clay Mathematics Institute, each carrying a one-million-dollar reward for a complete proof.
The effort ran for about 88 hours and consumed enormous computing resources: nearly three million messages exchanged between agents and roughly 130 billion output tokens generated, at an estimated cost approaching ten million dollars. The outcome, according to OpenAI, is that two of the four components required to fully validate the conjecture have been demonstrated, a meaningful step forward, though not yet enough to claim the full prize.
The announcement has not been universally welcomed in academic circles. Tristan Buckmaster, a mathematician at New York University, says he has spent months working on the very same problem alongside researcher Levent Alboğuz, using OpenAI’s own Codex tool as part of their process. He suggests that details of their progress reached the company before it went public with its own results, an implicit accusation that OpenAI drew on outside work without acknowledgment.
OpenAI has firmly denied accessing the researchers’ work ahead of its announcement, maintaining that its findings rest entirely on the internal capabilities of its new agent-based model.
Controversy aside, the episode points to a broader shift already underway: the growing use of large multi-agent systems to take on scientific challenges once considered the exclusive domain of the sharpest human minds. Whether this brute-force approach to computing power will genuinely reshape fundamental mathematical research, or simply stand as a headline-grabbing move in an increasingly fierce contest among AI labs, remains an open question, one that mathematicians, and rivals, will be watching closely.
