VisReason (Vision-Centric) - Spot the Difference: leaderboard

VisReason Spot the Difference is a perceptual-reasoning category of VisReason (2026): the model must find every difference between two scenes and localize each one with a bounding box. The score is accuracy in percent, higher is better.

Metric: Accuracy (%) on the Spot the Difference category (perceptual reasoning: find every difference between two scenes and localize each one with a bounding box) 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 2029. 20 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro (Preview)Low4.8#119
2GPT-5.2Thinking4.7#185
3GPT-5Low2.8#178
4O4 Mini2.7#368
5GPT-5 MiniLow2.2#364
6Qwen 3 VL 235B A22B (Thinking)1.6#459
7Qwen 2.5 VL 32B Instruct1.5#1062
8Qwen 3 VL 32B (Thinking)1.3#573
9Qwen 3 VL 8B Instruct1.1#1029
10GPT-4o0.8#751
11Keye-VL-1.5-8BThinking0.8#1720
12Qwen 3 VL 8B (Thinking)0.7#910
13MiMo-VL-7B-RL0.7#1155
14Gemini 2.5 Flash0.5#463
15Qwen 3 VL 30B A3B (Thinking)0.5#719

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=visreason-vision-centric-spot-the-difference · How It Works · Data refreshed daily, snapshot 2026-11-21.