VidPair-Halluc - Open-Ended Description: leaderboard
Metric: Open-ended description score (%) judged by GPT-4o on VidPair-Halluc adversarial video pairs (generated clips with near-identical backgrounds and different foreground semantics). Source: arxiv.org. Saturation forecast: Around May 2028. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 54.68 |
| 2 | Gemini 2.5 Flash | 50.23 |
| 3 | GPT-5 Mini | 49.33 |
| 4 | GPT-4o | 47.89 |
| 5 | Qwen 2.5 VL 7B Instruct | 45.07 |
Interactive version: theaggregate.ai/benchmark?slug=vidpair-halluc-open-ended-description · How It Works · Data refreshed daily, snapshot 2026-09-29.