Google releases Gemma 4 12B multimodal AI for laptops

Google has released a new artificial intelligence model designed to run efficiently on consumer laptops, marking a significant step in making advanced AI capabilities accessible without specialized hardware. The Gemma 4 12B model is an open-source, multimodal system that processes text, images, and audio directly on devices with as little as 16GB of memory.

The model represents a technical departure from conventional AI architecture. Most multimodal systems use separate, dedicated encoders to process images and audio before feeding that information to the core language model. Google's new approach eliminates these components. For vision, it employs a lightweight 35-million parameter module. For audio, it projects raw audio signals directly into the same dimensional space as text tokens. This "encoder-free" unified architecture reduces memory overhead and processing latency, allowing the model to run locally without the need for dedicated AI hardware.

Despite its smaller size of approximately 12 billion parameters, the company states the model delivers performance approaching its larger 26-billion parameter model in standard benchmarks. It supports a 256,000-token context window,...