Google's new open source Gemma 4 12B analyzes audio, video — and runs entirely locally on a typical 16GB enterprise laptop

Google released Gemma 4 12B, an open-weights 11.95B-parameter model under the Apache 2.0 license that can run locally on laptops with 16GB of VRAM or unified memory. The model supports audio, video, long-context processing, tool use, and step-by-step reasoning while avoiding separate encoders through a Unified architecture.

The release signals growing demand for capable AI that runs privately and offline, especially in enterprise settings where cost, latency, and data control matter. If performance holds up, compact local multimodal models could reduce reliance on cloud inference and accelerate edge AI adoption.