VURB - Long Video Understanding: leaderboard
Long-video split of VURB (Wei et al., 2026): 445 answer pairs to multiple-choice questions about videos of up to an hour from LVBench and the long subset of VideoMME, judged pairwise eight times with the answer order shuffled; a pair counts when the correct answer wins more than four judgments. Accuracy in percent over pairs with a valid verdict; higher is better.
Metric: Majority-vote pairwise accuracy (%; eight judgments with the response order shuffled, a pair counts when the chosen response wins more than four, over pairs with a valid verdict; 445 pairs from LVBench and VideoMME long). Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 63 | #172 |
| 2 | Qwen 3 VL 8B (Thinking) | 57.5 | #887 |
| 3 | Qwen 3 VL 8B Instruct | 57.1 | #982 |
| 4 | Qwen 3 VL 32B (Thinking) | 56.6 | #557 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vurb-long-video-understanding · How It Works · Data refreshed daily, snapshot 2026-11-12.