Google DeepMind Releases Gemma 4 12B for Laptops, Claiming 60%+ Quality Gain
Updated
Updated · developers.googleblog.com · Jun 3
Google DeepMind Releases Gemma 4 12B for Laptops, Claiming 60%+ Quality Gain
3 articles · Updated · developers.googleblog.com · Jun 3
Summary
Gemma 4 12B is now available across Google AI Edge tools on macOS, letting users run agentic, multimodal AI locally for coding, data analysis and text editing.
Google said the 12B model can generate and execute Python locally, build webpages, use tools and power a new Voice Edit feature in the fully offline AI Edge Eloquent app.
LiteRT-LM also adds a serve command, turning the model into an OpenAI-compatible local LLM server that can plug into standard SDKs, frameworks and coding extensions.
Google says Gemma 4 12B delivers a 60%+ jump in overall quality over prior models while keeping data on-device; earlier reports said it targets everyday laptops with about 16GB of RAM or VRAM.