Simon Willison used GPT-6 Sol at the max reasoning effort to generate an SVG of a pelican riding a bicycle from a one-line prompt. The public log records 18,227 tokens for that output (including 15,020 reasoning tokens) and an elapsed time of 2 minutes 35 seconds.
Suitable task: Turning a short creative description directly into an editable SVG draft.
Unsuitable task: Inferring general drawing quality, model rankings, or production costs from this example; it is a single creation log by the author, with no blind evaluation or repeated trials.
Applicable model version: The log identifies gpt-6-sol.
Applicable client, agent, or API: Simon Willison's llm CLI log; it does not specify the API, plugin version, or caller configuration.
Recommended reasoning effort and parameters: This example uses reasoning_effort: max. The source also records results for all six levels: none, low, medium, high, xhigh, and max. A single case cannot establish that max is more suitable for other tasks.
The following is the complete original prompt from the Gist's max result:
Generate an SVG of a pelican riding a bicycleThe complete SVG is available in the author's public Gist, and the sixth result, for the max level, can be viewed on the original result page. The original links are retained so readers can check the full code and image.
A Hacker News comment says the Gist shows results for all six GPT-6 Sol reasoning levels in order, with the max result at the end. A comparison grid separately shows pelican results for each GPT-6 and GPT-5.6 reasoning level. This article covers only the complete Sol max case.
The Gist records 16 input tokens, 18,227 output tokens (including 15,020 reasoning tokens), and 155,796 ms for this run. It provides no billing amount, API or plugin version, or independent quality score.
The original SVG has visible dimensions of 900 × 650. The image consists of daylight, grass, a teal bicycle, and a pelican riding it. The code is retained here as the generated result; this does not mean rendering has been verified in all browsers.
This is an author log for a single short prompt. It cannot establish the model's performance on other image tasks or support general conclusions about the benefits or costs of max.
GPT-6 Sol