QMFOLBench - Mathematics: leaderboard
Mathematics-topic subset of QMFOLBench (Zheng et al., 2026): 720 monadic first-order logic deduction tasks phrased about numbers, sets, statements and mathematicians, whose premises can contradict real mathematics, so the model must reason from the premises alone to label a conclusion True, False or Unknown. Accuracy in percent; higher is better.
Metric: Accuracy (%; 720 mathematics-topic tasks). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 98.75 | #91 (Gemini 3.1 Pro (Preview)) |
| 2 | GPT-5.4 (High) | 97.08 | #121 (GPT-5.4) |
| 3 | Qwen 3.5 27B (Thinking) | 93.06 | #358 (Qwen 3.5 27B) |
| 4 | Claude Sonnet 4.6 (Thinking) | 79.03 | #158 (Claude Sonnet 4.6) |
| 5 | DeepSeek V3.2 (Thinking) | 77.92 | #403 (DeepSeek V3.2) |
| 6 | DeepSeek V3.2 (Non-reasoning) | 67.5 | #403 (DeepSeek V3.2) |
| 7 | GPT-5.4 (Non-reasoning) | 62.64 | #121 (GPT-5.4) |
| 8 | Qwen 3 32B (Thinking) | 60.14 | #960 (Qwen 3 32B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=qmfolbench-mathematics · How It Works · Data refreshed daily, snapshot 2026-11-12.