Google brings Gemma 4 12B local AI workflows to laptops

6/3/2026

Gemma 4 12B powers local multimodal AI on laptops with Google AI Edge Run coding, dictation, and local model servers while keeping data on device

Google DeepMind has introduced Gemma 4 12B for local use on laptops through the Google AI Edge stack. The company says the open model is designed for agentic, multimodal tasks that can run on everyday machines while keeping data on device. The post highlights three main tools: Google AI Edge Gallery for local coding and data analysis on macOS, Google AI Edge Eloquent for ondevice dictation and editing, and LiteRTLM with a new serve command for running a local, OpenAIcompatible endpoint from the terminal. Google says the setup is intended to support fully local workflows such as generating charts from files, rewriting text with voice commands, and connecting standard developer tools to a local model server. It also points users to model cards and benchmarks for performance and memory requirements.