DatBench - Math & Logic: leaderboard
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around December 2026. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 8B (Thinking) | 43.3 |
| 2 | Qwen 3 VL 4B (Thinking) | 38.2 |
| 3 | Qwen 3 VL 8B Instruct | 35.9 |
| 4 | Qwen 3 VL 4B Instruct | 35.2 |
| 5 | GLM-4.1V-9B (Thinking) | 32.4 |
| 6 | Qwen 2.5 VL 7B Instruct | 27.9 |
| 7 | Gemma 3 4B (IT) | 15 |
| 8 | InternVL2-8B | 11.9 |
| 9 | InternVL2.5-2B | 9.9 |
Interactive version: theaggregate.ai/benchmark?slug=datbench-math-logic · How It Works · Data refreshed daily, snapshot 2026-09-25.