Rapid iteration by leading labs, particularly OpenAI’s GPT-5.6 series, has driven the 91.8% market-implied odds for an AI model reaching ≥90% on FrontierMath before 2027. As of September 2026, GPT-5.6 Sol already scores 83% on FrontierMath v2 Tier 4 and up to 89% on related leaderboards, following steep gains from sub-2% levels in late 2024 and post-v2 error corrections in June 2026 that lifted scores across the board. Continued scaling, improved reasoning scaffolds, and competitive releases from Anthropic and Google sustain trader expectations of further progress in the final months of 2026. Still, realistic headwinds include potential saturation on corrected problem sets, slower gains on the hardest Tier 4 holdouts, or unexpected benchmark revisions that could delay the threshold.