OpenAI's next-gen AI reportedly cracks 10 Fields Medal-level math problems

Chinese tech outlet cnBeta reports that OpenAI's next-generation AI has solved ten mathematical problems at a level associated with Fields Medal-class difficulty, according to a report published on August 2.

The report describes the achievement as a major milestone for OpenAI's upcoming model line, though it does not name a specific model or release date. The problems involved are described as among the hardest open questions in mathematics, the kind that have stumped researchers for years.

Fields Medal-level problems serve as a testing ground for frontier reasoning: unlike standard exam questions, they demand deep mathematical intuition and sustained multi-step reasoning. Solving ten of them at once would represent a significant leap over current-generation systems.

Math reasoning is widely treated as a proxy for general reasoning ability, which is why frontier labs compete so aggressively on hard-math evaluations. If the result holds up, OpenAI's next-generation model would sit at the front of that race.

For now, the report is based on media coverage and has not been officially confirmed by OpenAI. As with many capability claims, the details of how the problems were selected and verified will matter as much as the headline number.

The milestone also fits a broader industry pattern: each new frontier generation has pushed hard-math benchmarks to new heights, even as the gap between generations narrows and the problems get harder.

What to watch next: official confirmation from OpenAI, the benchmark or problem set used, and how rivals respond with their own next-generation models. If the claim holds, it could reset expectations for what reasoning models can do.