Epoch confirms GPT5.4 Pro solved a frontier math open problem

A frontier-level mathematics problem in hypergraph Ramsey theory has been solved for the first time by an AI. Researchers Kevin Barreto and Liam Price used OpenAI's GPT-5.4 Pro to develop a novel construction that improves lower bounds for the H(n) function, which relates to partitions in hypergraphs. The solution was confirmed by mathematician Will Brian, who called it "exciting" and noted it "eliminates an inefficiency in our lower-bound construction." Brian plans to write up the solution for academic publication, with Barreto and Price having co-authorship options.

Following this breakthrough, the research organization Epoch developed a standardized testing scaffold for their FrontierMath: Open Problems benchmark. Using this same framework, they found that other leading models—Anthropic's Claude Opus 4.6 (max), Google's Gemini 3.1 Pro, and another variant of OpenAI's GPT-5.4 (xhigh)—were also capable of solving the problem. This demonstrates that advanced reasoning and novel mathematical construction are now within reach of multiple state-of-the-art AI systems, moving beyond pattern recognition to genuine problem-solving in abstract domains.