Simon Willison has released version 0.33 of the llm-gemini plugin, bringing support for Google's newest models and compatibility with LLM 0.32. The update arrived on 13th August 2026, the same day Google launched Gemini 3.7 Flash.
New Models and Compatibility
The plugin now supports Gemini 3.7 Flash, gemini-3.6-flash, and gemini-3.5-flash-lite. It also adds two embedding models: gemini-embedding-2 and gemini-embedding-001. This marks the first llm-gemini release in some time, and it aligns the plugin with the latest version of LLM, the command-line tool for running language models.
LLM 0.32 introduced features that the plugin now takes advantage of. Users can view reasoning traces directly in their terminal. They can also enable server-side tools with a simple command pattern. For example, a user can run llm -m gemini-3.7-flash -T CodeExecution 'use python to calculate (factorial of 13) * 3' to execute code on Google's servers.
Thinking Efforts and a Notable Change
Gemini 3.7 Flash offers three thinking effort levels: high, medium, and low. The previous model, 3.6 Flash, had a minimal thinking effort option, but that has been removed in 3.7. Willison tested the new model by asking it to draw pelicans riding bicycles at each effort level.
The high-effort result, he said, is "pretty great." That image was rendered using Safari. Firefox and Chrome, however, displayed it differently. The bicycles appeared, but the pelicans were missing.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Browser Rendering Differences
The discrepancy comes down to how each browser handles SVG files. Willison noted that Safari is more tolerant of empty SVG <filter> elements than Firefox or Chrome. Those browsers skip the filter entirely, which causes the pelican to disappear from the final image.
It is a small quirk, but it highlights how model output can look different depending on the viewer. The underlying SVG code is the same in each case.
Sponsorship and Digest
Willison is offering sponsorship for a curated email digest of LLM developments. The cost is $10 per month. Subscribers receive regular updates on new tools, plugins, and model releases.
The digest is aimed at developers who want to stay current without scanning multiple sources. Sponsorship helps fund the ongoing maintenance of the LLM ecosystem.