Google Gemma 4 12B: Local multimodal AI for 16GB PCs
- Source
- VentureBeat
- Time
- 10:32 PM
- Weight
- 94/100
Google has released Gemma 4 12B, a new open-weights multimodal AI model designed to run locally on standard enterprise laptops with at least 16GB of RAM. Utilizing an "encoder-free" unified architecture, the model processes raw audio waveforms and visual data directly within its core backbone.
This design significantly reduces latency and memory overhead by eliminating the need for secondary processing modules, allowing for efficient offline use and enhanced data privacy for sensitive enterprise applications. Despite its compact size of approximately 12 billion parameters, the model features a 256K token context window and supports advanced capabilities such as native tool-use and a step-by-step reasoning mode.