Code2Video Benchmark
A benchmark that evaluates the ability of large models to generate video by writing single-file HTML/CSS/JavaScript code. Scoring focuses on visual and aesthetic dimensions such as interactivity, requirement compliance, composition, temporal consistency and production technique.
Timeline (last 90 days)
The Code2Video Benchmark released by Artificial Analysis sent 114 design requests to 27 mainstream models worldwide, requiring models to independently write a single HTML file to produce a video without using any external tools.
In the Code2Video Benchmark, every model's rendering success rate exceeded 96%.
The scoring stage of the Code2Video Benchmark is handled by Gemini 4 Argon, conducting pairwise comparisons across five dimensions: interactivity, requirement compliance, composition, temporal consistency and production technique.
The Code2Video Benchmark imposes a single-file HTML constraint: models cannot rely on external libraries, CDNs or asset resources, and must handle the layout system, animation loop and timeline coordination themselves.
In the Code2Video Benchmark, each question allows only one submission, and models cannot rework after seeing the finished video.
In the Code2Video Benchmark, the basic ability to generate code is no longer the main dimension distinguishing models; design aesthetics dimensions such as animation pacing, visual appeal and typography details have become the key to pulling ahead.
In the Code2Video Benchmark, high-priced models buy an aesthetic ceiling, while mid-range models buy an acceptable level of completion; the choice depends on the video's purpose (brand-level material or quick idea validation).
All Facts (7)
The Code2Video Benchmark released by Artificial Analysis sent 114 design requests to 27 mainstream models worldwide, requiring models to independently write a single HTML file to produce a video without using any external tools.
50%UnverifiedIn the Code2Video Benchmark, every model's rendering success rate exceeded 96%.
50%UnverifiedThe scoring stage of the Code2Video Benchmark is handled by Gemini 4 Argon, conducting pairwise comparisons across five dimensions: interactivity, requirement compliance, composition, temporal consistency and production technique.
50%UnverifiedThe Code2Video Benchmark imposes a single-file HTML constraint: models cannot rely on external libraries, CDNs or asset resources, and must handle the layout system, animation loop and timeline coordination themselves.
50%UnverifiedIn the Code2Video Benchmark, each question allows only one submission, and models cannot rework after seeing the finished video.
50%UnverifiedIn the Code2Video Benchmark, the basic ability to generate code is no longer the main dimension distinguishing models; design aesthetics dimensions such as animation pacing, visual appeal and typography details have become the key to pulling ahead.
50%UnverifiedIn the Code2Video Benchmark, high-priced models buy an aesthetic ceiling, while mid-range models buy an acceptable level of completion; the choice depends on the video's purpose (brand-level material or quick idea validation).
50%