OpenAI says its month-old model solved over 100 math problems

What OpenAI actually announced

The model’s results were presented alongside OpenAI’s new Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study. The group is intended to connect the company with mathematicians and the wider public as AI systems begin producing work with possible consequences beyond mathematics.

OpenAI says its collaboration is a “first step” toward answering two difficult questions: how AI can support mathematical understanding, and how the benefits of its capabilities can reach a broad audience.

The announcement leaves the central research claims underspecified. OpenAI has not said which problems the model solved, what method it used, or what the results mean for mathematics and science. Its published solution to a Navier–Stokes problem has already triggered sharp disputes among some researchers.

That disclosure also looks like a change in direction. Yakub Pachocki, OpenAI’s chief scientist, recently said the team had deliberately not optimized its system for mathematics, choosing instead to focus on recursive self-improvement. OpenAI addressed a millennium problem only after rumors that another team had solved it, partly with researchers from Anthropic.

Oversight without control

OpenAI says the advisory group will operate independently. Its members can:

OpenAI does not pay the participants. But the group cannot determine how quickly OpenAI advances its internal mathematics research. It may advise on whether and how results should be published, but not on whether the company should pursue those results or how fast it should proceed.

That distinction is the most revealing part of the arrangement. The group can help OpenAI interpret and communicate its research, while the company retains control over the research agenda and its pace.

The disagreement is about what mathematics is for

Timothy Gowers, a prominent mathematician and Fields Medal winner, is one of the group’s founders. He did not sign the open letter “A Serious Mismatch Between AI and Mathematics,” written by 25 Fields Medal winners, and explained his position in his blog.

Gowers agrees with much of the letter and believes mathematics is under threat. His disagreement concerns the purpose of the field:

Gowers says his Cambridge research project on automated theorem proving lost its purpose once large language models became capable enough to handle that work. Researchers had to accept what he called a “bitter lesson”: they knew this outcome was possible, but not how quickly it would arrive.

His larger concern is social rather than technical. People who might once have entered graduate school and become “guardians of the mathematical tradition” may decide that the effort is no longer worthwhile. Gowers says he was drawn to the profession by the dream of solving unsolved problems, especially famous ones. If that dream disappears, he is unsure what will replace it.

Funding creates a second vulnerability. Politicians may look at AI mathematics and conclude that human mathematicians are no longer necessary. Gowers argues that the field must quickly explain the value of a large community of human specialists, even if finding new theorem proofs is no longer part of their work.

The announcement is ahead of the evidence

My read is that OpenAI’s announcement is doing two jobs at once. It presents a rapid research result, and it creates a social framework for discussing that result before the company has disclosed enough for outsiders to evaluate it. Skeptics can reasonably see the timing as an attempt to reassure existing investors and attract new ones.

The quieter issue is not whether a model can generate a large number of solutions. It is whether the mathematical community can inspect, reproduce and build on those results without losing the human institutions that make mathematical knowledge durable. OpenAI’s advisory group can publish recommendations, but it cannot slow the company down or decide what research should happen. That leaves the company defining the pace of a transition whose costs may be paid by the people expected to sustain mathematics after the breakthrough.