Recent advances on Epoch AI’s FrontierMath: Open Problems benchmark have driven strong trader consensus that another official “Solved (AI)” designation will arrive soon. In mid-August 2026, Anthropic researchers used an unreleased Claude model to raise the proven lower bound on non-trivial zeta zeros on the critical line, with Lean formalization and expert validation; the same lab earlier credited Claude for solving the Hadamard matrix of order 668. OpenAI’s internal Astra system has simultaneously delivered multiple verified results, including Erdős problems and packing bounds. These outcomes reflect iterative gains in long-horizon reasoning, sub-agent coordination, and automated proof checking rather than isolated breakthroughs. With labs actively targeting additional moderately interesting open problems before year-end and Epoch AI updating its public list on a rolling basis, the next official classification could occur within weeks. Harder Millennium Prize targets remain distant.