WHOOPS Explanation of Violation — leaderboard
WHOOPS evaluates vision-language models on commonsense-defying images; this leaderboard snapshot tracks the explanation-of-violation human metric from the public WHOOPS full leaderboard.
Metric: Human Score (%). Source: huggingface.co. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Predicted Caption + GPT3 | 33 |
| 2 | BLIP2 FlanT5-XXL (Fine-tuned) | 27 |
| 3 | BLIP2 FlanT5-XL (Fine-tuned) | 15 |
| 4 | BLIP2 FlanT5-XXL (Zero-shot) | 0 |
Interactive version: theaggregate.ai/benchmark?slug=whoops-explanation-of-violation · How the rankings work · Data refreshed daily, snapshot 2026-07-22.