llm-gemini 0.33 adds Gemini 3.7 Flash support, reasoning traces

The llm-gemini 0.33 release marks a maintenance update to keep the plugin current with Google's Gemini model lineup. The addition of Gemini 3.7 Flash—Google's latest reasoning-capable model in the Flash family—reflects the ongoing evolution of the Gemini model family and gives developers access to the most recent inference capabilities via the command-line LLM tool.

Compatibility with LLM 0.32 is a key enabler: it unlocks reasoning-trace visibility and server-side tool execution. Reasoning traces are significant for developers who want to understand or debug how the model arrived at a conclusion, while server-side tools allow code execution (such as Python calculations) to happen within the model's context, reducing round-trips and enabling more complex workflows from the command line.

One minor friction point is the removal of the "minimal" thinking-effort option in 3.7 Flash, which slightly narrows the tuning options compared to 3.6 Flash and may affect workflows that relied on minimal-cost inference. Additionally, the SVG rendering inconsistency across browsers (Safari tolerating empty filter elements while Firefox and Chrome do not) suggests that developers generating SVG output via Gemini should test across browsers to catch display differences.