OpenAI claims model solved 100 open math problems in a month

OpenAI's announcement arrived alongside a new independent advisory group, the Advisory Group on Mathematics and Artificial Intelligence, based at the Institute for Advanced Study. The company frames the collaboration as a "first step" and says it needs outside guidance because math breakthroughs can have uses well beyond the field. The group operates independently in some respects: it is unpaid, its members choose who joins, they can offer advice without being asked and can publish recommendations. What it cannot do is decide how fast OpenAI moves — the group gets a say in how results are shared, not whether or how quickly they are produced.

The claim lands against a backdrop of criticism. The open letter "A Severe Misalignment of AI in Mathematics" argues that judging AI by the number of open problems it solves undermines conceptual understanding, which the signatories treat as the point of mathematics. Timothy Gowers, a founding member of OpenAI's group and a Fields Medalist, did not sign that letter. He agrees mathematics is in trouble, but draws the line differently: he describes a spectrum between mathematicians motivated mainly by solving problems and those motivated mainly by conceptual understanding, each seeing the other as a means. He also says his own Cambridge project on automated theorem proving lost its reason for existing once large language models got good enough.

The outcome hinges on questions OpenAI has left open: which problems the model actually solved, how, and what the results mean for math and science. Gowers's concern is less about the speed of solutions than about the people — whether those who would once have become "custodians of the mathematical tradition" still see a reason to, and whether funders decide human mathematicians are redundant. He expects capable LLMs to be publicly available within months, which suggests the pressure on those structures may build regardless of how the advisory group's recommendations land.