Google released Gemini 3.8 Flash today, introducing a new model variant with low, medium, and high thinking levels. The update arrives alongside a fix for async responses that previously failed to record the resolved model version. Simon Willison demonstrated the capabilities using his pelican illustrations, showing how the high thinking level produced a specific description of a cartoon bird on a bicycle compared to earlier outputs. He also highlighted the tool’s speed and cost efficiency when generating functional HTML and JavaScript code in under twenty seconds.
The practical value lies in the ability to build interactive web components directly from natural language prompts without extensive manual coding. Willison integrated the model into his markdown renderer to display HTML blocks within sandboxed iframes, allowing immediate visual feedback on generated content. This workflow reduces the friction between idea and implementation for developers needing quick prototypes.
* High thinking mode offers more detailed descriptive accuracy than previous versions.
* The model generated a working HTML page in thirteen seconds for one point eight cents.
* New rendering support allows HTML blocks to display in a secure iframe environment.
