🔍 Google introduces Gemma 4 12B, a multimodal open model that runs on a laptop

🔍 Google introduces Gemma 4 12B, a multimodal open model that runs on a laptop Google has unveiled Gemma 4 12B, its latest open-weight model with advanced reasoning, vision, and audio capabilities. Despite its size, it delivers performance close to larger Gemma models while running locally on just 16GB of VRAM and is released under the permissive Apache 2.0 license. The most interesting part is its new unified architecture. Most multimodal models rely on separate vision and audio encoders, which add memory overhead and latency. Gemma 4 12B largely removes them: 👁 Vision: A lightweight embedding module replaces the traditional vision encoder, allowing the LLM itself to handle visual understanding. 🎤 Audio: The audio encoder is removed entirely. Raw audio is projected directly into the same token space as text. The result is a smaller, faster, and more efficient multimodal model that can run on consumer hardware without sacrificing much capability. This could be a glimpse of where AI architectures are heading: fewer specialized components, more native multimodal reasoning. 🚀 Source. @aipost 🏴