Google has released 'Gemma 4 12B,' an AI model that can run on laptops, for free; it requires 16GB of VRAM to run.
On June 3, 2026, Google released its AI model ' Gemma 4 12B ' as an open model. Gemma 4 12B can run on systems with 16GB of VRAM or unified memory, and is being promoted as a high-performance model that can run even on laptops.
Introducing Gemma 4 12B
Gemma 4 12B: The Developer Guide - Google Developers Blog
https://developers.googleblog.com/gemma-4-12b-the-developer-guide/
The Gemma 4 series is an open model developed by Google using Gemini technology. On April 2, 2026, four models were released : the '31B' and '26B A4B' for high-performance PCs, and the 'E2B' and 'E4B' for edge devices such as smartphones. The newly released Gemma 4 12B is positioned as a model that 'fills the gap between the E4B and the 26B A4B,' and can run on PCs equipped with 16GB of VRAM or unified memory.
The benchmark results for 'Gemma 3 27B,' 'Gemma 4 12B,' and 'Gemma 4 26B A4B' are shown below. Gemma 4 12B recorded a score close to that of Gemma 4 26B A4B, which has a larger total number of parameters, and even surpassed Gemma 4 26B A4B in some tests.
The Gemma 4 12B is a multimodal model that supports image and audio input. While existing Gemma 4 series models converted images and audio using an encoder before processing, the Gemma 4 12B is designed to process them directly without encoding, successfully reducing latency and memory consumption.
The following video shows how Gemma 4 12B can perform text summarization and translation using voice commands.
Gemma 4 12B Demo: Native Audio Processing in Google AI Edge Eloquent - YouTube
Gemma 4 12B is released as an open model and is free to download. The base model 'gemma-4-12B', the chat-optimized 'gemma-4-12B-it', and the draft model for speculative decoding (MTP) 'gemma-4-12B-it-assistant' are available at the following link. The license is the Apache License 2.0.
google/gemma-4-12B · Hugging Face
https://huggingface.co/google/gemma-4-12B
google/gemma-4-12B-it · Hugging Face
https://huggingface.co/google/gemma-4-12B-it
google/gemma-4-12B-it-assistant · Hugging Face
https://huggingface.co/google/gemma-4-12B-it-assistant
Gemma 4 12B is already supported by LM Studio and Ollama , and can also be run using Google's local AI execution apps, Google AI Edge Gallery and Google AI Edge Eloquent . Note that macOS versions of Google AI Edge Gallery and Google AI Edge Eloquent are now available.
Related Posts:
