CoSimUE - Co-occurrence: leaderboard
Co-occurrence split of CoSimUE (He et al., Sichuan University, 2026): 338 yes/no question pairs on an MSCOCO image and a FLUX-edited copy that adds or swaps objects which usually co-occur with the scene, so a model leaning on context priors answers both images alike. Pair accuracy in percent (both answers right); higher is better.
Metric: Pair accuracy (%; 338 co-occurrence question pairs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 71.01 | #163 |
| 2 | Grok 4 | 70.41 | #306 |
| 3 | Gemini 2.5 Pro | 64.79 | #262 |
| 4 | Claude Sonnet 4.5 | 58.88 | #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-co-occurrence · How It Works · Data refreshed daily, snapshot 2026-11-12.