OpenAI’s GPT-6 Astra reached 97.6% on the corrected FrontierMath Tier 4 benchmark in September 2026, effectively saturating the set of 43 research-level problems that once stumped every model. Attention has now shifted to FrontierMath: Open Problems and the new Erdős collection of 68 still-unsolved conjectures, where Epoch AI verifies machine-checkable solutions in Lean or Python. Astra has already solved two Erdős problems, while other frontier models scored zero on the initial run. Trader sentiment tracks Epoch’s official classifications of newly solved open problems, upcoming model releases from OpenAI, Anthropic, and Google DeepMind, and any public announcements of additional verified solutions before key resolution dates. Progress on these genuinely unsolved problems remains slower and more uncertain than on the saturated tiers.