Epoch AI’s FrontierMath benchmark now contains approximately 50 unsolved research-level mathematics problems in its Open Problems track. AI systems have solved exactly three of them, representing a 6% success rate. Over roughly 18 months between late 2024 and mid-2026, this rate climbed from near zero to 30-40% across the broader benchmark. The Open Problems track, launched in early 2026, specifically tests AI capabilities on questions that no mathematician has yet resolved. One of the three solved problems involved a Ramsey-style hypergraph problem in combinatorics.
Source: Read the original article
