Recent AI progress on Epoch AI’s FrontierMath: Open Problems benchmark, which tracks genuinely unsolved research-level math problems with verifiable solutions, continues to shape trader views. GPT-6 Astra’s September 2026 launch delivered near-saturation scores on the corrected Tier 4 set and contributed to at least one additional Open Problems solve marked by Epoch AI in mid-September, following earlier AI-assisted resolutions such as the Ramsey hypergraph problem and the 2-adic absolute Galois group. Competitive dynamics favor OpenAI and Anthropic models that leverage extended inference budgets and human-AI collaboration, while Epoch’s policy updates on crediting such results lower barriers to official classification. Key upcoming catalysts include further runs with the newest frontier models and any new problem markings before resolution dates, though exact timelines remain uncertain given verification and publication requirements.