AutoHallusion: leaderboard
AutoHallusion — automatically generated hallucination benchmark for vision-language models; probes object/scene conflicts that induce hallucinated answers.
Metric: Overall Accuracy. Source: huggingface.co. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4V | 66 | |
| 2 | Gemini Pro Vision | 51.4 | |
| 3 | miniGPT4 | 51 | |
| 4 | LLaVA-1.5 | 44.5 | |
| 5 | Claude 3 | 37.1 |
Interactive version: theaggregate.ai/benchmark?slug=autohallusion · How It Works · Data refreshed daily, snapshot 2026-10-20.