OpenAI’s GPT-6 Astra has driven recent trader sentiment by posting 97.6% on FrontierMath v2 Tier 4, the hardest slice of Epoch AI’s research-level math benchmark, while human-AI collaborations have produced additional verified solutions on the separate Open Problems track. Epoch AI recently classified an Apéry-style irrationality problem as solved under its new “human plus AI” category, following earlier marks on hypergraph and Galois group questions, though none matched the exact autonomous form originally targeted. Competitive pressure remains intense, with Claude Fable variants close behind on corrected tiers and Google DeepMind models showing strength on related formal-reasoning tasks. Key near-term catalysts include Epoch’s next leaderboard refresh, potential new Erdős-set evaluations, and any official classification of further open problems before year-end resolution windows.