VidPair-Halluc - Multi-Choice F1: leaderboard
VidPair-Halluc (2026) probes hallucination in video-language models with adversarial pairs of generated videos that share a near-identical background but differ in foreground actions, attributes or event order. This board is multiple-choice F1 in percent for assessing and ranking clip-level descriptions; higher is better.
Metric: Multiple-choice F1 (%) for assessing and ranking clip-level descriptions on VidPair-Halluc adversarial video pairs (generated clips with near-identical backgrounds and different foreground semantics). Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 67.32 | #266 |
| 2 | GPT-5 Mini | 64.82 | #348 |
| 3 | Gemini 2.5 Flash | 62.44 | #453 |
| 4 | Qwen 2.5 VL 7B Instruct | 61.5 | #1833 |
| 5 | GPT-4o | 59.15 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vidpair-halluc-multi-choice-f1 · How It Works · Data refreshed daily, snapshot 2026-11-13.