OpenAI published AI-generated full or partial solutions Tuesday to more than 370 outstanding mathematical problems, including some long treated as grand challenges in the field.
OpenAI said it used an unreleased internal AI model and said the model took about three hours of computing time on average to reach each solution. The company said the batch includes full or partial results for many of the problems mathematicians have considered among the most important in the field.
The results come weeks after OpenAI said it had used the same kind of unreleased internal model to solve the Navier-Stokes equations, one of the seven Millennium Prize problems for which the Clay Mathematics Institute offers a $1 million award. In the latest batch, OpenAI said it made progress on three other Millennium Prize problems, but did not fully solve them.
Mathematicians split over what the announcement means. Some were enthusiastic, seeing new areas for research. Others argued that AI companies are turning long-standing mathematics problems into a stage for model demonstrations and marketing points.
UCLA mathematician Terence Tao is among the prominent members of the field who have criticized the way OpenAI and other AI companies have targeted major math problems to test new models. Dan Litt, a professor of mathematics at the University of Toronto, said he was excited by the results and called them 'great for mathematics,' but he also warned that a perception that AI has 'solved math' could hurt funding and discourage young mathematicians.
Litt said it is important that society continue to support human mathematical expertise if the field is going to benefit from the progress AI has made. OpenAI said the work was done with its internal model, and that the model required about three hours of computing time on average for each solution.
There is a broader business lesson here for every AI company racing to claim the next breakthrough. The market rewards speed, but mathematics is still a discipline built on rigor, verification, and human judgment. When a model is used to produce a vast cache of results, the question is not just what it can do. It is who controls the rules, who gets the credit, and whether the rush for attention crowds out the people doing the underlying work.
That tension is exactly why this story matters beyond academia. Capital rewards clear rules, but public claims around frontier AI often arrive faster than the institutions meant to verify them. If AI firms want legitimacy, they will need more than scale and computing time. They will need trust.
