Rapid progress on Epoch AI’s FrontierMath benchmark—where top models reached roughly 52% accuracy on Tiers 1–3 by mid-2026 after earlier o-series results around 25%—has driven trader consensus toward an 83% implied probability of ≥90% scores before 2027. Continued scaling of large language models, improved reasoning chains, and tool use are expected to close the remaining gap within the next six to eighteen months, consistent with historical saturation patterns on other math benchmarks. Key near-term catalysts include anticipated GPT-5.5-class and successor releases plus any public updates from Anthropic or Google DeepMind through year-end 2026, though Tier 4’s greater difficulty and potential evaluation changes introduce modest uncertainty.