Release: llm-gemini 0.34
- Added the
gemini-3.8-flashmodel for Gemini 3.8 Flash, with support for low, medium, and high thinking levels. #146- Fixed an issue where asynchronous responses could not record the parsed model version. Thanks to Charlie Tonneslan. #137
Google released Gemini 3.8 Flash today, along with 3.8 Flash Cyber, which is available only to “trusted defenders.”
Here are pelican images generated using the high, medium, and low thinking levels. Here’s the result generated at the “high” level:

For comparison, here is the same set of pelican images generated with Gemini 3.7 Flash.
One thing I really appreciate about Gemini Flash is that it’s fast, inexpensive, and particularly capable at tasks involving HTML and JavaScript. On a whim, I tested it by entering the prompt “Build me something cool in HTML,” and it actually built something cool in HTML! It took 13 seconds and cost 1.8 cents.
<video controls=“controls” preload=“none” poster=“https://static.simonwillison.net/static/2026/cosmic-gemma.jpg” width=“854” height=“370” style=“display: block; width: 100%; height: auto;”
<source src="https://static.simonwillison.net/static/2026/cosmic-gemma.mp4" type="video/mp4" />Your browser does not support HTML5 video.Visit the demo page to see another project I built using Gemini 3.8 Flash.
My markdown-svg-renderer tool accepts a Gist URL containing Markdown and renders the Markdown, including SVG in fenced code blocks correctly.
I used Gemini 3.8 Flash, along with my very basic llm-coding-agent coding agent plugin, to add HTML support to it. Any HTML blocks in the Markdown are now rendered in a sandboxed iframe. Here’s the complete transcript.