VisReason (Vision-Centric) - Cue Insight: leaderboard
Metric: Accuracy (%) on the Cue Insight category (conceptual reasoning: combine textual and visual clues in an image to derive the answer) 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 January 2027. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) (Low) | 68.5 |
| 2 | GPT-5 Mini (High) | 60.4 |
| 3 | GPT-5 Mini (Medium) | 55.9 |
| 4 | GPT-5 (Low) | 53.2 |
| 5 | O4 Mini | 46 |
| 6 | GPT-5 Mini (Low) | 45.1 |
| 7 | GPT-5.2 (Thinking) | 43.2 |
| 8 | Qwen 3 VL 8B Instruct | 34.3 |
| 9 | Qwen 3 VL 235B A22B (Thinking) | 30.6 |
| 10 | GPT-5 Nano | 29.7 |
| 11 | Qwen 3 VL 30B A3B (Thinking) | 28.4 |
| 12 | GPT-4o | 27.1 |
| 13 | Qwen 3 VL 32B (Thinking) | 27 |
| 14 | Qwen 3 VL 8B (Thinking) | 25.2 |
| 15 | Qwen 2.5 VL 32B Instruct | 21.6 |
Interactive version: theaggregate.ai/benchmark?slug=visreason-vision-centric-cue-insight · How It Works · Data refreshed daily, snapshot 2026-10-07.