All benchmarks

Artifact generation benchmarks

Compare artifact quality across text models

Pelican riding a bicycle

Why we chose this prompt: The classic, as a 32×32 grid instead of SVG. There is no circle element to lean on for the wheels, only 1,024 cells, so the picture shows whether a model can draw or only remembers the shapes.

Prompt

Pixel art of a pelican riding a bicycle, drawn on a 32x32 grid with flat colors and no anti-aliasing

Sort by
Author
Toughest critic
Favicon for openai
GPT-5.6 Sol
−1.1 vs median
Biggest fan
Favicon for google
Gemini 3.1 Pro Preview
+0.7 vs median