QMFOLBench - Mathematics: leaderboard

Metric: Accuracy (%; 720 mathematics-topic tasks). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview) (High)98.75
2GPT-5.4 (High)97.08
3Qwen 3.5 27B (Thinking)93.06
4Claude Sonnet 4.6 (Thinking)79.03
5DeepSeek V3.2 (Thinking)77.92
6DeepSeek V3.2 (Non-reasoning)67.5
7GPT-5.4 (Non-reasoning)62.64
8Qwen 3 32B (Thinking)60.14

Interactive version: theaggregate.ai/benchmark?slug=qmfolbench-mathematics · How It Works · Data refreshed daily, snapshot 2026-09-25.