Vibe check
A pelican on a bicycle
One prompt, one API key, several models. Same request body every time — only the model id changes.
Generate an SVG of a California brown pelican riding a bicycle. The bicycle must have spokes and a correctly shaped bicycle frame. The pelican must have its characteristic large pouch, and there should be a clear indication of feathers. The pelican must be clearly pedaling the bicycle. The image should show the full breeding plumage of the California brown pelican.
- Endpoint
- /v1/chat/completions
- Temperature
- 1
- Max output tokens
- 32,000
- System prompt
- none
- Reasoning effort
- medium / low / high
How these were made
Every drawing on this page came from the same request, sent to 58 models through one CrossModel key. Nothing else changed between them — not the endpoint, not the SDK, not the auth header.
curl https://api.crossmodel.ai/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer cm-YOUR_KEY" \
-d '{
"model": "anthropic/claude-fable-5",
"messages": [
{
"role": "user",
"content": "Generate an SVG of a California brown pelican riding a bicycle. The bicycle must have spokes and a correctly shaped bicycle frame. The pelican must have its characteristic large pouch, and there should be a clear indication of feathers. The pelican must be clearly pedaling the bicycle. The image should show the full breeding plumage of the California brown pelican."
}
],
"temperature": 1,
"max_tokens": 32000
}'Whose test this is
The test is not ours: the prompt below is v2 of Simon Willison's pelican-riding-a-bicycle benchmark, quoted verbatim — he has run it against every new model since 2024, and revised it in November 2025 to demand a correctly built bicycle and a California brown pelican in breeding plumage. We reproduce it here through our own gateway; the parameters are ours.
Reasoning effort is set explicitly rather than left at each model’s default, which ranges from medium to xhigh — effort is the biggest lever on this task, so leaving it unset would have made the runs incomparable in the one dimension that matters most. Models with no reasoning mode get no such setting. The output budget above is the one every model started from: a model whose own ceiling is lower was clamped down to it, and two that spent the entire budget reasoning without drawing anything were given a larger one. Either way the card says which figure that model actually ran with.
This is not a benchmark
There is no score here, and nothing on this page is measured. A drawing tells you how a model handles an open-ended task on one particular day; it does not rank anything. For published, source-attributed scores, see the benchmark section on each model page.
Generated 2026-09-14. These are stored results — nothing is generated when you open this page.