Rapid progress on FrontierMath, Epoch AI’s suite of unpublished research-level math problems, underpins the 80% market-implied odds for a model reaching 90% before 2027. GPT-5.6 Sol currently leads at 89% on the main set and 83% on v2 Tier 4 following the June 2026 error corrections that lifted scores across the board. Earlier GPT-5.5 variants crossed 50% on Tiers 1-3 in April 2026 after starting near zero in late 2024, reflecting gains from larger models, extended reasoning, and tool use. With roughly 16 months remaining, continued scaling, new frontier releases, and test-time compute advances position labs to close the final gap, though benchmark noise on the hardest tier and any further revisions introduce modest uncertainty for traders.