animation2code benchmark
For best compatibility, please view this dashboard in a Chrome browser.

Zero-shot video (or image-frame) → code results on the test set, across commercial and open-source models.

Each output is tagged with A = appearance similarity and T = temporal similarity; higher is better for both. Click a video to inspect its code.

6572 of 214

ground truth

CSS Waves

model outputs

GPT-5.4

A 0.91T 0.28

Claude Sonnet 4.6

A 0.76T 0.27

LLaMA 4 Scout

A 0.60T 0.24

ground truth

CSS animated waves

model outputs

GPT-5.4

A 0.92T 0.36

Claude Sonnet 4.6

A 0.78T 0.36

LLaMA 4 Scout

A 0.64T 0.27

model outputs

GPT-5.4

A 0.81T 0.28

Claude Sonnet 4.6

A 0.87T 0.39

LLaMA 4 Scout

A 0.61T 0.19

model outputs

GPT-5.4

A 0.68T 0.14

Claude Sonnet 4.6

A 0.78T 0.15

LLaMA 4 Scout

A 0.74T 0.14

model outputs

no output

Qwen3-VL-8B-Instruct

A T

GPT-5.4

A 0.89T 0.32

Claude Sonnet 4.6

A 0.88T 0.29

LLaMA 4 Scout

A 0.52T 0.25

model outputs

GPT-5.4

A 0.98T 0.19

Claude Sonnet 4.6

A 0.99T 0.11

LLaMA 4 Scout

A 0.49T 0.11

ground truth

Water Drop

model outputs

no output

Qwen3-VL-8B-Instruct

A T

GPT-5.4

A 0.83T 0.29

Claude Sonnet 4.6

A 0.86T 0.30

LLaMA 4 Scout

A 0.64T 0.26

model outputs

GPT-5.4

A 0.79T 0.27

Claude Sonnet 4.6

A 0.89T 0.23

LLaMA 4 Scout

A 0.40T 0.00