Epoch AI confirms GPT-5.4 Pro solved a decades-old hypergraph partitioning problem for the first time, verified by mathematician Will Brian. The solution, developed by Kevin Barreto and Liam Price, improves lower bounds for H(n) using a recursive algorithm that generates hypergraphs with no isolated vertices. Brian praised the AI’s approach for eliminating inefficiencies in existing constructions, calling it “a perfect mirror” of human upper-bound methods. The breakthrough comes after rigorous testing via Epoch’s FrontierMath scaffold, where GPT-5.4 Pro outperformed models like Gemini 3.1 Pro and Opus 4.6.

The problem, rooted in Ramsey theory, asks for hypergraphs with specific partition constraints. GPT-5.4 Pro’s algorithm produces valid witnesses for H(n) ≥ c*k_n (c > 1), a mathematical first. Brian noted the solution’s potential to spur follow-on research, with Barreto and Price eligible for coauthorship. Other models solved the problem too, but GPT-5.4 Pro’s efficiency and scalability stood out, completing tasks in under 10 minutes on standard hardware.

Mathematicians surveyed rated the problem’s difficulty as requiring 1–3 months of human effort. GPT-5.4 Pro’s success—achieved via prompt engineering and iterative refinement—highlights AI’s growing role in tackling abstract challenges. The solution’s publication in a specialty journal is anticipated, with implications for combinatorial optimization.

Key entities: Epoch AI, GPT-5.4 Pro, Will Brian, Kevin Barreto, Liam Price. Critical figure: 10x faster than human estimates. Primary keyword: AI solved math problem. Secondary keywords: Frontier Math, hypergraph algorithms, Ramsey theory, computational verification.