VisReason (Vision-Centric) - Deductive Reasoning: leaderboard
Metric: Accuracy (%) on the Deductive Reasoning category (conceptual reasoning: derive the logically entailed conclusion from the visual and textual information) of VisReason, answers scored by format (regular-expression match, GPT-5-mini judge for open-ended answers, IoU above 0.5 for boxes), zero-shot with chain-of-thought prompt templates, one per answer format; rows not shaded gray in the paper use explicit reasoning (GPT-5 and Gemini 3 Pro at low reasoning effort); higher is better. Source: arxiv.org. Saturation forecast: Around July 2027. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) (Low) | 53.7 |
| 2 | GPT-5.2 (Thinking) | 48.9 |
| 3 | Qwen 3 VL 235B A22B (Thinking) | 45.2 |
| 4 | GPT-5 Mini (Medium) | 44.4 |
| 5 | GPT-5 Mini (High) | 41.4 |
| 6 | Qwen 3 VL 32B (Thinking) | 39.2 |
| 7 | GPT-5 (Low) | 38.8 |
| 8 | GPT-5 Mini (Low) | 37.7 |
| 9 | Qwen 3 VL 30B A3B (Thinking) | 37.3 |
| 10 | O4 Mini | 36.9 |
| 11 | Qwen 3 VL 8B Instruct | 34.1 |
| 12 | Gemini 2.5 Flash | 34 |
| 13 | Qwen 3 VL 8B (Thinking) | 34 |
| 14 | GPT-4o | 32.1 |
| 15 | GPT-5 Nano | 24.6 |
Interactive version: theaggregate.ai/benchmark?slug=visreason-vision-centric-deductive-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-07.