Skip to content

Benchmark

VBench-Long subject consistency

Scores are percentages.

Models measured
9
Scores recorded
9
Sources
1
IQ weight
0.10

Score distribution

96.6%
Mean
97%
Median
0.896%
Std dev
95.8%
10th percentile
97.4%
90th percentile
94.6%
Min
94.6%97.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.10 for video-generation models, measured over 8 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Video 01· VBench data docs97.5%
2Kling V1.6· VBench data docs97.4%
3Veo 3· VBench data docs97.4%
4Hunyuan Video2025-05-22· VBench data docs97.2%
5Mochi 1· VBench data docs97%
6LTX Video5s 768×512· VBench data docs96.6%
7Sora· VBench data docs96.2%
8Vidu Q12025-04-17· VBench data docs95.8%
9Vidu Q4· VBench data docs94.6%

Data credits: VBench (Vchitect), MIT (licence)