6/5 Revision: 16GB RAM → VRAM.
https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/
Ars Technica article
Google’s new Gemma 4 12B model is designed to run on any laptop with 16GB of RAM
https://arstechnica.com/google/2026/06/googles-new-gemma-4-open-ai-model-is-sized-for-your-laptop/
Some content was translated from Gemini 3.1 Flash Lite and edited.
Last April, Google announced Gemma 4, releasing two mobile-optimized versions (E2B and E4B) and two models for more professional work (26B Mixture of Experts and 31B Dense). There was a gap in the middle range, which the newly released 12B fills.
Gemma 4 12B offers significantly better performance than the mobile version but doesn't require very expensive AI accelerators to run locally.
Google says that this new 12B model is unique in its ability to run on many consumer laptops without sacrificing quality. All you need is a computer with 16GB VRAM to run this 12 billion parameter model.
Gemma 4 12B uses about half the total memory of Gemma 4 26B MoE, and Google claims it delivers nearly equivalent performance based on benchmarks.
Gemma 4 12B incorporates a newly designed multi-token prediction (MTP) drafter that utilizes unused processing cycles to calculate future tokens. This results in improved speed and efficiency.
While Google has previously released optional MTP versions for other Gemma 4 models, this is the first time an MTP-enabled model has been released as standard.
The Gemma 4 12B model is available through LM Studio and the Google AI Edge Gallery app.
https://lmstudio.ai/models/gemma-4
https://developers.google.com/edge/gallery
