OpenAI’s early October release of 722 AI-generated math manuscripts from an unreleased internal model, including claimed advances on the Unique Games Conjecture and other long-standing problems, has sharply lifted trader expectations for its systems on FrontierMath open problems. GPT-6 Astra variants have already secured multiple verified solves on Epoch AI’s Erdős and Tier 4 sets, outpacing rivals in volume and compute efficiency. Anthropic’s Claude Fable series and Google DeepMind’s AlphaProof Nexus have each delivered peer-reviewed or benchmark-credited results on elliptic curves and Erdős problems, keeping competitive pressure high. With FrontierMath Tier 4 now largely saturated and new Erdős problems still open, the next official Epoch classification will likely hinge on which lab’s latest large language model or agent swarm produces the first Lean-verified proof on a remaining high-notability entry.