**Recent AI breakthroughs on research-level math problems are driving near-certain market odds that Epoch AI will officially recognize at least one more solved FrontierMath: Open Problems entry by late 2026.** Epoch AI launched its FrontierMath: Open Problems benchmark in late July 2026, featuring 50 genuinely unsolved research questions vetted by mathematicians. At launch, a handful had already been solved (including a Ramsey-style hypergraph problem and one on the 2-adic absolute Galois group), and the market explicitly excludes those. Since then, multiple high-profile advances have occurred: OpenAI’s Astra model announced solutions to 10 longstanding open problems in early August 2026; Anthropic’s Claude Fable 5 produced a counterexample to the Jacobian conjecture (a century-old algebraic geometry question) around mid-July to early August; and agents such as Aletheia have resolved or advanced several Erdős problems, with independent verification. These results stem from improved reasoning scaffolds, tool use, and inference-time scaling in frontier large language models. Trader sentiment reflects the accelerating pace—models from OpenAI, Anthropic, and Google are now routinely producing verifiable outputs on Tier 4/research-grade problems that previously stumped experts. Key catalysts ahead include continued model iterations, new Epoch AI evaluations, and human expert confirmation of AI-generated proofs. Resolution depends solely on Epoch AI’s official classification on its site, not third-party claims, which adds a modest verification lag but has not slowed the recent trend. The market-implied odds (high 90s for December) align with the demonstrated momentum in AI mathematical reasoning capabilities.