The llm-gemini plugin has been updated to version 0.33, extending compatibility to several Gemini Flash models and adding two embedding models.
Key changes
- Official support for Gemini 3.7 Flash, plus support for gemini-3.6-flash and gemini-3.5-flash-lite.
- Two new embedding models added: gemini-embedding-2 and gemini-embedding-001.
- Upgraded for compatibility with LLM 0.32, enabling reasoning traces and the ability to enable server-side tools.
Example of enabling server-side tools
With LLM 0.32 compatibility, server-side tools can be invoked using a pattern such as:
llm -m gemini-3.7-flash -T CodeExecution \
'use python to calculate (factorial of 13) * 3'
Rendering note and cross-browser behavior
The author tested Gemini 3.7 Flash by generating images — for example, pelicans riding bicycles at different "thinking effort" levels (high, medium, low). The "minimal" option that existed in 3.6 Flash has been removed in 3.7.
A practical issue was observed in rendering the generated SVG: Safari displayed both the pelican and the bicycle, whereas Firefox and Chrome displayed the bicycle but the pelican was missing. This discrepancy was attributed to Safari being more tolerant of empty SVG <filter> elements than the other browsers.
Summary
Version 0.33 focuses on adding model and embedding support and improving integration with LLM 0.32 features. The included generation examples also highlight that visual outputs can vary across browsers, which should be considered when validating graphical results.
