Simon Willison has updated his llm-gemini plugin to support new reasoning levels for Google's Gemini 4 models. The latest release, llm-gemini 0.38.1, addresses an oversight where Gemini 4 Carbon, Hydrogen, and Neon were only configured for low, medium, and high reasoning, while Hydrogen and Neon also offer an 'xhigh' level. This 'xhigh' setting for Gemini 4 Hydrogen significantly increases output tokens and cost but yields competitive results, as demonstrated by a pelican generation example. Willison also notes the confusing pricing structure for the Gemini 4 series, particularly with the introduction of Gemini 4 Neon, which has further complicated the lineup and cost comparisons. AI
IMPACT Updates to LLM plugins and model reasoning levels can improve performance and cost-efficiency for developers.
RANK_REASON Simon Willison's blog posts detail updates to his llm-gemini plugin and observations on Google's Gemini 4 models, rather than a direct release from a frontier lab.
- Argon
- Carbon
- .claude
- Datasette
- Gemini 4
- Gemini 4 Neon
- Hydrogen
- Pyodide
- Python
- Qwen 3.8-27B
- Gemini 4 Hydrogen
- llm-gemini
- Simon Willison
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →