Summary & Key Takeaways ​

• Simon Willison conducted a creative benchmark for LLMs.
• The challenge involved generating an SVG of "an armadillo in fishnet tights jaywalking on Mars."
• He tested models like Claude Opus 5.5, GPT-6.1-Sol, Gemini 3.8-Flash, and Mistral Large 4.
• The exercise highlights the generative and creative capabilities of different LLMs.

Our Commentary ​

I love this kind of benchmarking. Forget the dry academic papers; asking an LLM to generate "an armadillo in fishnet tights jaywalking on Mars" is a far more entertaining and often insightful way to gauge its creative and generative capabilities. It's a fun, messy way to see what these models can really do.