Gemma 4 is Google DeepMind's most advanced open-weight model family, released on April 2, 2026 under the Apache 2.0 license. The flagship 31B model beats systems 20x larger while running locally on a single GPU, supporting 140+ languages with native multimodal capabilities including text, vision, and audio understanding. Available in multiple sizes for deployment anywhere from mobile devices to workstations, Gemma 4 represents a pareto frontier in open AI models with top-tier arena scores and full compatibility with frameworks like transformers, llama.cpp, MLX, and WebGPU.