VisReason (Vision-Centric) - Open-Ended: leaderboard
Metric: Accuracy (%) on the 309 open-ended questions (judged correct or not by GPT-5-mini against the reference answer) of VisReason, pooled over the 10 reasoning categories, 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 January 2027. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) (Low) | 72.5 |
| 2 | GPT-5 (Low) | 61.5 |
| 3 | GPT-5 Mini (High) | 60.2 |
| 4 | GPT-5 Mini (Medium) | 55.3 |
| 5 | GPT-5.2 (Thinking) | 52.1 |
| 6 | GPT-5 Mini (Low) | 50.5 |
| 7 | O4 Mini | 48.5 |
| 8 | Qwen 3 VL 235B A22B (Thinking) | 43.7 |
| 9 | GPT-5 Nano | 29.8 |
| 10 | Qwen 3 VL 32B (Thinking) | 29.8 |
| 11 | Qwen 3 VL 8B Instruct | 28.6 |
| 12 | Qwen 3 VL 30B A3B (Thinking) | 25.6 |
| 13 | Qwen 3 VL 8B (Thinking) | 25.1 |
| 14 | Qwen 2.5 VL 32B Instruct | 23.9 |
| 15 | Gemini 2.5 Flash | 12 |
Interactive version: theaggregate.ai/benchmark?slug=visreason-vision-centric-open-ended · How It Works · Data refreshed daily, snapshot 2026-10-07.