On October 26, 2026, HKU Business School announced that Seedance 2.0 Pro ranked first in generation performance and sixth in safety and responsibility in its AI video benchmark. The two placements came from separate assessments, so the top generation result did not carry over to safety.
The Artificial Intelligence Evaluation Lab (AIEL) at HKU Business School compared 12 models for generation performance. The rankings describe that benchmark’s results.
What the benchmark measured
HKU evaluated text-to-video generation and text-and-image-to-video generation, which combines a text prompt with a reference image. It assessed six dimensions: visual continuity, physical plausibility, aesthetic quality, instruction alignment, audio quality, and audio-visual matching.
Together, those dimensions cover how consistently a video holds together from frame to frame, whether its action seems physically plausible, how it looks, how well it follows instructions, and how its audio sounds and matches the visuals.
Seedance, Veo, and Sora visual scores
In the study’s visual-only comparison, Seedance 2.0 Pro scored 3.84 out of 5, ahead of Veo 3.1 at 3.47 and Sora 2 Pro at 3.27. SCMP reported these scores for the three models.
| Model | Visual score (out of 5) |
| Seedance 2.0 Pro | 3.84 |
| Veo 3.1 | 3.47 |
| Sora 2 Pro | 3.27 |
These figures describe visual performance in this benchmark. They are separate from HKU’s safety-and-responsibility ranking.
A separate assessment for safety
HKU placed Seedance 2.0 Pro sixth for safety and responsibility. Its safety assessment used prompts covering misinformation, illegal content, violence and explicit content, bias and discrimination, and copyrighted material.
The generation ranking and safety placement therefore describe different parts of HKU’s evaluation: video-generation performance across the stated dimensions, and responses to prompts in high-risk categories.
