Rapid progress in large language model mathematical reasoning has driven the 79.5% market-implied odds that an AI system will reach ≥90% on FrontierMath before 2027. Top models, including recent OpenAI GPT-5 variants, have already posted 89% on legacy snapshots and 87–88% on the corrected Tier 4 v2 released by Epoch AI in June 2026, narrowing the gap to the threshold amid continued scaling, inference-time compute gains, and domain-specific fine-tuning. Traders view the remaining four-plus months of 2026 as sufficient for incremental releases or training runs to close the final points, consistent with historical benchmark trajectories where performance jumps accelerated near saturation. Upcoming catalysts include further model launches from OpenAI, Anthropic, and Google DeepMind, alongside Epoch AI’s ongoing evaluations that could confirm or refute the crossing before year-end.