A ridiculous prompt.
A revealing picture.

What is the Pelican Bicycle Test? An unexpectedly tricky request: draw a pelican that is actually riding a bicycle.

“Generate an SVG of a pelican riding a bicycle.”

Knowing the nouns is the easy part.

An AI model can produce a long beak, a bird-shaped body, two circles, and a frame. That doesn’t mean it has drawn a bird riding a bicycle. The body must relate to the seat; the legs must relate to the pedals; the wings or body must plausibly meet the handlebars. Those relationships make the prompt useful.

Why SVG?

SVG asks a model to turn visual reasoning into explicit shapes and coordinates. It makes both the image and the underlying technical artifact inspectable. Pelican Score safely converts that artifact to a fixed PNG so every submission gets the same viewport and background.

What the test can tell you.

It offers a small window into instruction following, spatial composition, recognizable objects, and mechanical relationships. It also creates a shareable, slightly absurd way to compare the outputs you actually generated.

What it cannot tell you.

Pelican Score is an experimental community benchmark, not an official scientific benchmark. Scores depend on a limited rubric and an imperfect AI judge. They do not measure every capability of a model, and the uploader’s model label is self-reported.

The home page’s hand-drawn pelican is a demo illustration, not a submitted model result. Public rankings contain actual completed submissions judged by a real configured vision provider.