QMFOLBench - University: leaderboard

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

Top models

#ModelScore
1Gemini 3.1 Pro (Preview) (High)99.86
2GPT-5.4 (High)98.19
3DeepSeek V3.2 (Thinking)96.67
4Qwen 3.5 27B (Thinking)92.92
5Claude Sonnet 4.6 (Thinking)84.72
6DeepSeek V3.2 (Non-reasoning)79.72
7Qwen 3 32B (Thinking)65.42
8GPT-5.4 (Non-reasoning)61.25

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