Google DeepMind Ships Gemma 4 12B — Encoder-Free Multimodal Model With Native Audio That Runs on a 16GB Laptop

Google The Keyword·high signal

Released June 3, the 11.95B-param open-weights model reads text, image, audio, and video with no separate vision or audio encoder, fitting in ~16GB (Q4 ~6.7GB) under Apache 2.0 with a 256K context window. It nears the larger 26B MoE on standard benchmarks at under half the memory footprint, with day-one support across Transformers, llama.cpp, MLX, vLLM, SGLang, and Unsloth. Encoder-free native audio at this size is a genuine local-inference unlock for builders.