Overview

We built FrontierMath: Open Problems (FM:OP) to help track AI math capabilities. As of July of 2026, AI systems have already made significant autonomous contributions to research mathematics, e.g., on the unit distance problem. These results function as an existence proof: AI can make highly significant contributions to research mathematics. But several questions remain, and we think FM:OP can help answer them.

First, not all results in mathematics are of equal significance. With FM:OP we’ve tried to distinguish between several tiers on this dimension. To do so we have relied on the opinions of mathematicians, most notably our editorial board — you can read their thoughts on the benchmark and how to understand progress on it.

Second, FM:OP can provide something of a “denominator” for AI contributions to math. The best AI math results to-date are indeed impressive, but it is difficult to tell how prevalent such results are among problems of significant interest. As a curated problem set, FM:OP serves as a stable reference point for tracking future AI results.

Third, even the most impressive AI contributions to math have, so far, consisted of applying known techniques. It would mark a significant step forward for AI if it were able to develop new theory from whole cloth in order to solve a problem. We cannot know in advance which (if any) FM:OP problems might require such innovation: that will only become clear from post hoc analysis of solutions. But we expect the benchmark to help us monitor for the emergence of such capabilities.

See our FAQ to learn more about the benchmark, or explore the individual problems.