CoSimUE - Similarity: leaderboard
Similarity split of CoSimUE (He et al., Sichuan University, 2026): 174 yes/no question pairs on an MSCOCO image and a copy with small random FLUX edits that turn objects into look-alikes, with the correct answer flipped between the two images. Pair accuracy in percent (both answers right); higher is better.
Metric: Pair accuracy (%; 174 similarity question pairs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4 | 79.31 | #306 |
| 2 | GPT-5 | 75.86 | #163 |
| 3 | Gemini 2.5 Pro | 72.41 | #262 |
| 4 | Claude Sonnet 4.5 | 62.64 | #244 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cosimue-similarity · How It Works · Data refreshed daily, snapshot 2026-11-12.