Simon Willison (software developer and open-source advocate) has released version 0.30 of his llm-gemini plugin, broadening the horizons for developers using his popular command-line tool. This update provides critical access to Google's newest model releases, including the lightweight yet efficient Gemini 3.1 Flash Lite Preview. By integrating these models, users can leverage rapid processing speeds for simpler tasks while maintaining the flexibility of a terminal-based workflow.

The update also introduces support for the Gemma 4 family, specifically the 26B and 31B versions. These are "open" models, meaning the mathematical weights are available for users to run on their own hardware or private servers. The 31B variant represents a significant mid-sized option, balancing computational requirements with sophisticated reasoning. This allows students and researchers to experiment with high-performing language models without relying on massive enterprise infrastructure.

The ecosystem serves as a vital bridge between diverse AI providers and the developer community. By abstracting the complexities of different cloud interfaces, the plugin ensures that switching between models is as simple as changing a single command. This democratization of access is essential for those exploring agentic engineering—the practice of using AI as a proactive assistant to complete complex, multi-step software tasks.

Simon Willison (software developer and open-source advocate) has released version 0.30 of his llm-gemini plugin, broadening the horizons for developers using his popular command-line tool. This update provides critical access to Google's newest model releases, including the lightweight yet efficient Gemini 3.1 Flash Lite Preview. By integrating these models, users can leverage rapid processing speeds for simpler tasks while maintaining the flexibility of a terminal-based workflow.

The update also introduces support for the Gemma 4 family, specifically the 26B and 31B versions. These are "open" models, meaning the mathematical weights are available for users to run on their own hardware or private servers. The 31B variant represents a significant mid-sized option, balancing computational requirements with sophisticated reasoning. This allows students and researchers to experiment with high-performing language models without relying on massive enterprise infrastructure.

The ecosystem serves as a vital bridge between diverse AI providers and the developer community. By abstracting the complexities of different cloud interfaces, the plugin ensures that switching between models is as simple as changing a single command. This democratization of access is essential for those exploring agentic engineering—the practice of using AI as a proactive assistant to complete complex, multi-step software tasks.