- Published on
Llm-gemini 0.33
- Authors
- Name
- DeepHorus
A new version of the llm-gemini plugin, version 0.33, has been released, adding support for today's Gemini 3.7 Flash model along with several other AI models. The update, announced on 13th August 2026, also includes compatibility with gemini-3.6-flash, gemini-3.5-flash-lite, and two embedding models: gemini-embedding-2 and gemini-embedding-001. This marks the first release in a while for the plugin, which is widely used by developers interacting with Google's Gemini models through the command-line LLM tool.
The plugin has been upgraded to work with LLM 0.32, a key improvement that enables users to see reasoning traces from the models. Additionally, it now supports server-side tools, allowing users to trigger features like code execution directly from the command line. For example, the command 'llm -m gemini-3.7-flash -T CodeExecution' can run Python calculations without additional setup.
The release notes highlight a notable test involving Gemini 3.7 Flash generating images of pelicans riding bicycles at various thinking efforts. The author of the release notes experimented with high, medium, and low thinking settings, noting that the 'minimal' option available in 3.6 Flash has been removed in 3.7. The high-effort image was described as 'pretty great,' while lower-effort versions were not shown.
An interesting technical issue emerged during these tests: the generated pelican image rendered correctly in Safari but failed to display fully in Firefox and Chrome. The problem appears to be caused by Safari's more tolerant handling of empty SVG filter elements, which the other browsers ignore, resulting in a missing pelican while the bicycle remains visible. This highlights ongoing cross-browser compatibility challenges with AI-generated SVG content.
For developers and AI enthusiasts, this release is significant because it brings the latest Gemini models into a familiar command-line workflow. The ability to see reasoning traces and use server-side tools could streamline debugging and automation tasks. However, the removal of the 'minimal' thinking effort in Gemini 3.7 Flash may disappoint users who preferred faster, lighter responses.
Looking ahead, the llm-gemini plugin is expected to keep pace with Google's rapid model releases. The inclusion of multiple new embedding models suggests a focus on retrieval and search applications. As browser rendering of AI-generated graphics continues to vary, developers may need to test outputs across different browsers before deploying them in production environments.
