AI model scores ≥ 90% on FrontierMath Benchmark before 2027?

Prediction markets put the probability at 14%: AI model scores ≥ 90% on FrontierMath Benchmark before 2027. Currently, markets see this as unlikely (14% YES). Elad Gil Says AI Will Hit 1% Of U.S.

What’s Happening

A prediction market currently assigns a 14% probability to the event that an AI model scores ≥ 90% on the FrontierMath Benchmark before 2027, with an overwhelming 86% of participants betting against the milestone. FrontierMath, a rigorous mathematical reasoning test designed to challenge AI systems, has become a key barometer for advanced model capability. The low confidence in this outcome comes amid a flurry of major model releases: OpenAI launched GPT-5.5 on April 24, 2026, touting gains in agentic coding and scientific research, while Chinese firm DeepSeek unveiled its V4 model the same day, intensifying competition. These launches highlight rapid progress, yet the market suggests that even the most advanced systems remain far from mastering the frontier-level mathematics required for a 90% score on the FrontierMath Benchmark. [MLQ.ai, Apr 24]

The significance of the ai model scores ≥ 90% on frontiermath benchmark target extends beyond a single test, as it represents a potential inflection point in AI reasoning capabilities. Investor Elad Gil, in a thesis published on April 20, 2026, argued that AI has grown from near-zero to 1% of U.S. GDP and that the window for AI founders to exit is narrowing, underscoring the economic stakes of such breakthroughs. Meanwhile, geopolitical tensions are reshaping the landscape: China’s DeepSeek and Huawei chips are challenging U.S. dominance, with analysts noting that the AI race is no longer one-sided. A model achieving 90% on FrontierMath would signal a leap in abstract problem-solving, potentially accelerating both commercial applications and national security concerns. [Forbes, Apr 26]

Looking ahead, the path to a 90% score on the FrontierMath Benchmark before 2027 faces both technical and competitive hurdles. Tencent previewed its Hy3 base model on April 23, 2026, emphasizing product integration over raw benchmark performance, a strategy that may prioritize real-world utility over mathematical extremes. The iPhone 18 is rumored to launch in early 2027 with 12GB of RAM and an A20 chip, hinting at hardware advances that could support more capable on-device AI, but not necessarily the specialized reasoning needed for FrontierMath. With the 2027 deadline approaching, the market’s 86% NO vote reflects a consensus that current architectures—even those behind GPT-5.5 and DeepSeek V4—are unlikely to bridge the gap to near-perfect mathematical reasoning within the next eight months. [PGurus, Apr 26]

Traded on Polymarket — $57K Volume

Lower-volume market on Polymarket ($57K). Wider spreads expected — enter with limit orders and be aware of slippage risk. Currently 14c YES.