Google releases Gemma 4 12B open-source multimodal model for local deployment
Key Questions
What are the key features of Google's Gemma 4 12B model?
Gemma 4 12B is an encoder-free open-source multimodal model supporting 256K context, MTP, function calling, and thinking mode, with audio and video analysis capabilities.
What hardware is required to run Gemma 4 12B locally?
The model runs on typical 16GB RAM laptops, making it suitable for edge deployment and easily fitting in 32-64GB VRAM setups alongside the larger 26B variant.
Where can developers access Gemma 4 12B?
It is available on Kaggle Models and released by Google DeepMind as an open-source model for local deployment and experimentation.
New Gemma 4 12B from Google DeepMind is encoder-free, supports 256K context, MTP, function calling, and thinking mode. Runs on 16GB RAM, making it accessible for edge deployment and complementing the 26B variant. Open-source, fits in 32-64GB VRAM easily.
Sources (2)
Updated Jun 4, 2026