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.

7380 of 214

model outputs

GPT-5.4

A 0.90T 0.31

Claude Sonnet 4.6

A 0.92T 0.35

LLaMA 4 Scout

A 0.40T 0.10

model outputs

GPT-5.4

A 0.93T 0.32

Claude Sonnet 4.6

A 0.94T 0.33

LLaMA 4 Scout

A 0.93T 0.31

model outputs

GPT-5.4

A 0.92T 0.34

Claude Sonnet 4.6

A 0.92T 0.32

LLaMA 4 Scout

A 0.84T 0.28

model outputs

GPT-5.4

A 0.94T 0.31

Claude Sonnet 4.6

A 0.89T 0.28

LLaMA 4 Scout

A 0.91T 0.27

model outputs

GPT-5.4

A 0.88T 0.31

Claude Sonnet 4.6

A 0.92T 0.28

LLaMA 4 Scout

A 0.83T 0.28

model outputs

GPT-5.4

A 0.86T 0.30

Claude Sonnet 4.6

A 0.90T 0.31

LLaMA 4 Scout

A 0.82T 0.30

ground truth

Google+ Jingle

model outputs

GPT-5.4

A 0.81T 0.26

Claude Sonnet 4.6

A 0.82T 0.26

LLaMA 4 Scout

A 0.67T 0.22

ground truth

Nice spinny stuff

model outputs

GPT-5.4

A 0.85T 0.32

Claude Sonnet 4.6

A 0.93T 0.31
no output

LLaMA 4 Scout

A T