OpenAI’s GPT-6 Astra has saturated Epoch AI’s FrontierMath Tier 4 benchmark at 97.6 percent, shifting trader focus to the harder FrontierMath: Open Problems set of genuinely unsolved research questions. Recent releases, including hundreds of AI-generated proofs for long-standing conjectures across number theory, topology, and logic, have accelerated expectations that frontier large language models will soon produce verifier-accepted solutions on additional Epoch-tracked problems. Competitive pressure from Anthropic’s Claude models and ongoing Lean formalization efforts add momentum, while verification delays, occasional withdrawals of flawed proofs, and the benchmark’s strict Epoch-only resolution criteria create uncertainty around exact timing. Traders watch upcoming model updates and Epoch evaluations for the next official “solved” designation.