What exactly does the Pelican Test ask a model to do?
Generate a single self-contained HTML file that shows a pelican riding a bicycle, with all styles, SVG and images inlined in that one file.
This page reads the same live leaderboard query as the public star leaderboard, isolates qwen4-max, and lists four reference models with identical fields. It computes no rank and does not treat visitor ratings as a capability score.
Every model below comes from the same live leaderboard query: published runs, rating count and the average of signed-in visitor ratings. The order is this fixed comparison list, not a ranking.
| Model | Published runs | Ratings | Average rating |
|---|---|---|---|
qwen4-max | 1 | 0 | Unrated |
claude-opus-5 | 1 | 1 | 5.0 / 5 |
gpt-6-astra | 1 | 1 | 5.0 / 5 |
qwen3.8-max | 1 | 1 | 4.0 / 5 |
glm-5.3 | 1 | 1 | 4.0 / 5 |
Live query: no ratings yet (averageStars: null), so this model is unrated — not zero stars.
A real zero-star average is shown as 0.0 / 5; unrated only means no one has rated the model yet.
Captured from the public isolated preview endpoint. The generated HTML is not injected into this page, and the preview CSP and sandbox are unchanged.

The common task is to generate a single self-contained HTML file showing a pelican riding a bicycle. The star average comes only from signed-in visitor votes.
Generate a single self-contained HTML file that shows a pelican riding a bicycle, with all styles, SVG and images inlined in that one file.
Only signed-in visitors can rate, subjectively, from 0 to 5 stars. Each account can rate each model once, and 0 stars is a valid negative vote.
No. It is the average of subjective visitor votes. It is not normalized for thinking strength or other provider effort settings, and it is not a capability score.
No. The page lists qwen4-max and four reference models side by side with the same fields in a fixed comparison order. It does not calculate, display or imply a rank.
No. A null average means the model is unrated. A 0.0 / 5 average is shown only when real visitor votes gave zero stars.
The page and the public star leaderboard use the same server-side leaderboard query. This page selects five model keys; a missing entry or a failed query is shown as data unavailable, never backfilled.
Run the Pelican Test with your own API key. Your API provider may charge for usage. Ratings are signed-in visitor opinions, once per account per model.
Live figures come from the public leaderboard query. The screenshot is a historical snapshot of the named run, not a live iframe.