Bottom line: Gemma 4 from Google DeepMind brings advanced AI agent capabilities to end devices. The open-source model enables autonomous planning and decision-making in over 140 languages – entirely without specialization or fine-tuning.
Google DeepMind has released Gemma 4 – a new open-source model that enables advanced AI capabilities directly on users’ end devices. The system, licensed under Apache 2.0, supports autonomous decision-making, multi-step planning, and is available in over 140 languages.
Google’s AI Edge Team has unveiled a new generation of open-source models with Gemma 4 that expands the boundaries of on-device AI applications. The model enables developers to deploy intelligent agents and autonomous AI systems directly on end devices – without requiring additional fine-tuning.
Gemma 4’s core capabilities include multi-step planning, autonomous decision-making, offline code generation, and audio and image processing. This makes it possible to realize complex AI applications beyond simple chatbots. The model was specifically developed for a global audience and supports over 140 languages, which significantly improves international deployment capability.
Under the Apache 2.0 license, the model is now available to developers to support them in building powerful edge AI applications.