OpenAI’s GPT-6 Astra launch on September 3, 2026, and its reported 97.6% on FrontierMath v2 Tier 4 underscore accelerating large language model reasoning capabilities, following GPT-5.4 Pro’s verified solution of a 2019 hypergraph problem in March and a second open problem resolved by mid-2026. Epoch AI’s Open Problems track, now expanded with dozens of research-level challenges, relies on rigorous verifier checks before marking any solution official, creating a high bar that favors well-resourced labs. Competitive pressure from Anthropic’s Claude Fable series and Google models, combined with OpenAI’s lead in benchmark scores and test-time compute, drives trader consensus toward additional solves by year-end. Key near-term catalysts include further Epoch AI evaluations of new frontier systems and potential model releases that could extend the recent streak of AI-discovered proofs or counterexamples.