66.7%
#1 of 10
VBench-2.0 total score of Veo 3, the highest of 10 models in early 2026; Vidu Q1 62.7%
What the chart shows
VBench-2.0 scores video generators on how faithfully their output follows reality across five dimensions: human fidelity, creativity, controllability, physics and commonsense. Veo 3 led the 10 models listed with a total score of 66.7%, ahead of Vidu Q1 (62.7%) and ToMoviee 2.0 (61.8%).
The other seven models are bunched between 53.4% (CogVideoX-1.5) and 60.2% (Wan2.1). Sora at 480p resolution scored 58.4% and Kling 1.6 scored 59.0%.
Key numbers
| Measure | Value |
|---|---|
| Veo 3 | 66.72% |
| Vidu Q1 (2025-04-17) | 62.7% |
| ToMoviee 2.0 | 61.78% |
| Kling 1.6 | 59.0% |
| CogVideoX-1.5 | 53.35% |
How to read it
- Snapshot of the VBench leaderboard (Vchitect) in early 2026.
- VBench-2.0 combines vision-language and language model judges with specialised detectors, checked against human preference labels.
- Dates in some model names identify the version evaluated.
Source: VBench Leaderboard, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.3.7: VBench-2.0: total score. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion