PriVE-Bench: leaderboard
PriVE-Bench (2026) shows vision-language models counterfactual images whose visible content contradicts a familiar prior, such as an unusual part count or logo layout, in five domains, and a text-only judge classifies each answer as correct, prior-consistent or other. Score is counterfactual accuracy macro-averaged over the five domains, in percent; higher is better.
Metric: Counterfactual accuracy (%; macro-averaged over the five domains) on paired original and counterfactual images in five domains, counterfactual-only input, answers classified by a text-only judge (GPT-4o-mini for closed models, Qwen3-8B for open models); temperature 0, reasoning off or minimal. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (Minimal) | 46.7 | #167 (GPT-5) |
| 2 | Claude Sonnet 4 | 46.6 | #400 |
| 3 | Qwen 3 VL 32B Instruct | 40.3 | #585 |
| 4 | Gemma 3 12B (IT) | 39.7 | #1837 |
| 5 | Qwen 3 VL 8B Instruct | 31.6 | #981 |
| 6 | Gemma 3 27B (IT) | 30.5 | #1289 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=prive-bench · How It Works · Data refreshed daily, snapshot 2026-11-13.