**OpenAI’s rapid iteration on the GPT-5.6 series has driven the 89.5% market-implied probability for Yes.** As of late August 2026, GPT-5.6 Sol leads the FrontierMath v2 leaderboard at 89% on Tiers 1–3 (295 corrected research-level problems) and 83% on the harder Tier 4 set, just one point shy of the 90% threshold. The June 2026 v2 update—after Epoch AI found errors in 42% of the original problems—lifted scores across the board while preserving relative rankings, accelerating the perceived timeline. Earlier 2026 models sat in the low-to-mid 50% range on v1; the corrected benchmark and heavier reasoning variants have compressed progress dramatically. With only four months remaining before 2027 and frontier labs releasing frequent updates, traders view another incremental gain to 90% as the base case before year-end, though benchmark saturation and evaluation variance introduce modest residual uncertainty.