**OpenAI’s GPT-6 Astra launch in early September 2026 drove recent trader sentiment by saturating FrontierMath Tier 4 at 97.6–98% accuracy and delivering the first verified solutions on Epoch AI’s genuinely open research problems.** The model solved two of 68 Erdős problems in official runs (five across attempts) while competitors such as GPT-5.6 Sol and Claude Fable 5.1 scored zero on the same set. Epoch’s shift toward harder open-problem collections and Lean-verified benchmarks has amplified expectations that additional solutions will be formally recognized soon. Traders are watching Epoch’s next evaluation cycles, potential GPT-6.1 or Claude Opus 5.5 runs, and any new model releases that could accelerate progress on the remaining open problems before year-end deadlines.