OpenAI’s October 6 release of 722 AI-generated mathematical manuscripts—spanning algebra, number theory, topology, and claims touching Millennium Prize problems—has sharply reinforced trader expectations that its internal frontier model will deliver the next verified FrontierMath open-problem solution. GPT-6 Astra already leads Epoch AI’s FrontierMath v2 Tier 4 leaderboard at 97.6% and solved two of 68 Erdős problems in early testing, while competing systems from Anthropic and others trail significantly on the same private, Lean-verified benchmarks. Traders are watching Epoch AI’s ongoing open-problems evaluations and any formal recognitions of new solutions, which remain the sole resolution trigger; recent scale and verification speed favor OpenAI but leave room for surprise contributions from other labs if evaluation access expands.