**OpenAI’s unreleased internal frontier model has driven recent trader sentiment by releasing 722 manuscripts on October 6–7, 2026, advancing or solving 372 open mathematical problems across number theory, algebra, topology, and related fields.** This follows its September claim of progress on the Navier-Stokes equations, one of the Clay Millennium Prize problems. GPT-6 Astra has already saturated FrontierMath Tier 4 (v2) at 97.6% on Epoch AI’s corrected benchmark of 43 research-level problems, while legacy tiers show GPT-5.6 variants leading at 80–89%. Epoch AI’s separate FrontierMath: Open Problems set tracks genuinely unsolved conjectures (such as Erdős problems formalized in Lean), with resolution depending solely on Epoch’s official classification of new solutions. Competitive pressure from Anthropic’s Claude models and Google DeepMind’s agentic systems adds to the pace, though verification of large-scale AI outputs remains a key uncertainty. Near-term catalysts include further Epoch evaluations, proof formalizations, and any additional model releases before year-end deadlines.