All benchmarks

Artifact generation benchmarks

Compare artifact quality across text models

Sketch benchmarks

Pelican on a bike

Why we chose this prompt: Simon Willison's informal benchmark for text models, run verbatim since late 2024. Text models cannot draw, but they can write code and SVG is code. The task is hard on purpose. Bicycle frames are difficult to remember, pelicans are difficult to draw, and pelicans cannot ride bicycles. Each model gets the prompt with no examples or style guidance, so the picture reflects its own judgment.

Prompt

Generate an SVG of a pelican riding a bicycle

Sort by
Author
Toughest critic
Favicon for google
Gemini 3.8 Flash
−1.3 vs median
Biggest fan
Favicon for openai
GPT-5.6 Sol
+0.6 vs median