Recent announcements from OpenAI and Anthropic have accelerated trader focus on AI's 2026 math capabilities. In September, OpenAI used thousands of AI agents on an unreleased frontier model to produce a Lean-verified proof of finite-time blow-up for a forced Navier-Stokes variant, one of the Clay Millennium Prize Problems, following months of multi-agent experimentation. October saw OpenAI release 722 manuscripts resolving 377 open problems across number theory, algebra, and topology, with partial claims touching the Riemann hypothesis and Yang-Mills existence, while Anthropic agents formalized Fermat's Last Theorem in 11 days and contributed to refutations of long-standing complexity hypotheses. These results, many Lean-certified, highlight competitive scaling of large language model agents and formal verification tools, though full peer review and Clay recognition remain pending. Traders are watching November-December verification workshops and potential follow-up releases that could expand solved problem sets before year-end.