VNU-Bench - Cross-source Evidence Integration: leaderboard
VNU-Bench (Liu et al., University of Florida, 2026) question type in which a multimodal model must combine textual and visual evidence from several outlets into one reasoning chain, resolving conflicts, given a group of YouTube news videos on one event. Each video is given as 4 selected frames and a cleaned transcript cut to 1,000 characters; zero-shot at temperature 0. Accuracy in percent on 4-option multiple-choice questions; higher is better.
Metric: Accuracy (%; question type T6, 4-option multiple choice). Source: arxiv.org. Saturation forecast: Around June 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 60.08 | #244 |
| 2 | Gemini 2.5 Pro | 60.02 | #262 |
| 3 | Qwen 2.5 VL 7B Instruct | 49.22 | #1842 |
| 4 | InternVL3-8B | 49.06 | #1695 |
| 5 | Gemini 2.5 Flash | 45.96 | #452 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vnu-bench-cross-source-evidence-integration · How It Works · Data refreshed daily, snapshot 2026-11-12.