DiCoBench - Instance Commonality: leaderboard

Metric: Accuracy (%) on the instance-commonality task (re-identify one physical object across very different scenes) of DiCoBench (765 multi-image samples at near-2K resolution): 5-option multiple choice (four candidate image pairs or instructions plus a no-visible-commonality option, chance 20%), answer letter matched exactly, VLMEvalKit settings at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 18 models tracked.

Top models

#ModelScore
1Qwen 2.5 VL 7B Instruct74.6
2Gemini 3 Pro72.7
3O4 Mini71.8
4GPT-4.1 Mini70.9
5GPT-566.4
6Qwen 3.5 35B A3B64.6
7Gemma 3 12B (IT)50
8Gemma 3 27B (IT)47.3
9Gemini 3 Flash45.5
10Qwen 2.5 VL 32B Instruct45.5
11GPT-4o28.2
12GPT-4.122.7

Interactive version: theaggregate.ai/benchmark?slug=dicobench-instance-commonality · How It Works · Data refreshed daily, snapshot 2026-09-29.