GoKawiilNew benchmark expands Simon Willison's pelican-bicycle SVG test with 30 prompts
A community-built site has extended Simon Willison's informal LLM benchmark—asking models to draw an SVG of a pelican riding a bicycle—into a set of 30 drawing prompts tested across 2025 and 2026 model runs. The 2026 batch currently covers prompts 1 through 10, with images displayed unedited exactly as each model generated them.
GoKawiil's interpretation of the reporting above, not reported fact.
Informal creative benchmarks like this one have become a popular, low-cost way for developers to compare how AI models handle spatial reasoning, abstraction and humor rather than just raw text accuracy. Expanding the test to 30 varied prompts and tracking results year over year gives a rougher but more visual signal of how quickly image-generation and code-generation capabilities in LLMs are improving.
- The site expands Willison's original pelican-bicycle SVG test into 30 distinct drawing prompts.
- Results are tracked across separate 2025 and 2026 model runs for comparison.
- The 2026 dataset is still in progress, currently showing outputs for prompts 1 to 10.
Source: gally.net — This Site Was Created Claude Fable, 2026-09-14
Published there as: “Show HN: Pelican-bicycle alternatives (updated for 2026)”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.