AI model scores ≥ 90% on FrontierMath Benchmark before 2027?
As of July 14, 2026, the market gives “AI model scores ≥ 90% on FrontierMath Benchmark before 2027?” a 88% chance of YES.
| YES odds | 88% |
|---|---|
| NO odds | 12% |
| Volume | $108,036 |
| Closes | December 31, 2026 |
Updated July 14, 2026 · Live data from Polymarket
About this market
Rapid gains in frontier AI reasoning capabilities explain the 79.5% market-implied odds for an AI model reaching 90% on FrontierMath before 2027. Top models have climbed from under 2% accuracy in late 2024 to 25-48% on Epoch AI’s tiers by mid-2026, driven by improved chain-of-thought, tool integration with code interpreters, and larger-scale training. Recent releases such as OpenAI’s o-series and GPT-5 variants, alongside Google’s Gemini updates, demonstrate accelerating progress on unpublished research-level problems. Traders expect continued scaling and architectural refinements through 2026 developer cycles to close the remaining gap, though benchmark updates and potential plateaus introduce residual uncertainty.