VisuLogic: leaderboard
Visual logic reasoning benchmark requiring models to perform logical deduction from visual inputs without relying on language shortcuts.
Metric: Overall Accuracy (%). Source: visulogic-benchmark.github.io. Status: years away from saturation. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 235B A22B (Thinking) | 34.4 |
| 2 | O3 | 29.5 |
| 3 | Gemini 2.0 Pro (Preview 02-05) | 28 |
| 4 | InternVL3-78B | 27.7 |
| 5 | InternVL2.5-78B | 27.3 |
| 6 | InternVL3-38B | 27.1 |
| 7 | GPT-4o | 26.3 |
| 8 | Qwen 2.5 VL 7B Instruct | 26 |
Interactive version: theaggregate.ai/benchmark?slug=visulogic · How It Works · Data refreshed daily, snapshot 2026-09-05.