Mistral Large 4
Simon Willison's linkblog entry on the release of Mistral Large 4, built around a Hacker News comment arguing that standard benchmarks are saturated. The commenter demonstrates this by running the identical whimsical prompt, Generate an SVG of an armadillo in fishnet tights jaywalking on Mars, through Claude Opus 5.5, GPT-6.1 Sol, Gemini 3.8 Flash, and Mistral Large 4 via Willison's llm CLI tool, then compares the rendered SVGs at each model's default reasoning level using his markdown-svg-renderer tool.
What it claims
The entry is a one-line Hacker News quote from user wren6991 ("The benchmark is saturated. Frontier models are tested with an armadillo in fishnet tights jaywalking on Mars.") and Willison's response: running that prompt through four current models. It makes no capability claim about Mistral Large 4. The page gives no parameter count, context length, pricing, licence or benchmark figure for the model. The title is the only thing that ties the entry to the release.
How to use it
The page's practical value is the exact llm model identifiers for reproducing the comparison. The Mistral model uses a mistral/ prefix and the other three do not, which suggests it is served through a separate Mistral plugin for llm rather than built in. The page does not show the plugin install step.
llm -m claude-opus-5.5 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars'
llm -m gpt-6.1-sol 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars'
llm -m gemini-3.8-flash 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars'
llm -m mistral/mistral-large-4 'Generate an SVG of an armadillo in fishnet tights jaywalking on Mars'All four runs used each model's default reasoning level. No reasoning or effort flags were passed. The outputs are viewed through tools.simonwillison.net/markdown-svg-renderer, which takes a ?url= parameter pointing at the saved output. The full URL is truncated in the page text.
Limits
- The fetched text contains no outputs, rankings or commentary on which model did better. Any comparison is in the rendered SVGs behind the truncated link, so the reader has to open that link to see results.
- The runs used one prompt, one sample per model and default reasoning levels, which differ per model. This is an informal spot check, not an evaluation. It does not support the "benchmarks are saturated" framing with any data.
- An engineer evaluating Mistral Large 4 for adoption will need to find size, licence, cost and benchmark details from another source, because this page has none of them.
Organisations named here
Technologies named here
- LLM
- Claude Opus 5.5
- markdown-svg-renderer
- GPT-6.1 Sol
- Gemini 3.8 Flash
- Mistral Large 4
- Mistral
- Gemini
- Claude
Also named here
- Simon Willison
Related in models


