Rapid progress in large language model reasoning has driven the 90% market-implied odds for an AI system reaching 90% on FrontierMath before 2027. Top models such as GPT-5.6 Sol and Claude Fable 5 now score 82–89% on current tiers of the Epoch AI benchmark, up sharply from under 30% in early 2025, reflecting gains from scaled training, specialized post-training, and iterative improvements on unpublished research-level math problems. Traders view the remaining gap as bridgeable within months given typical release cycles and competitive pressure among labs. Key near-term catalysts include potential new model drops or benchmark updates through year-end, though exact timelines for Tier 4 saturation remain subject to technical hurdles and evaluation changes.