Google DeepMind Introduces Gemma 4 Open Model Family

Google's DeepMind has introduced **Gemma 4**, a new generation of open AI models designed for advanced reasoning, code generation, and multimodal understanding. The models are released under the Apache 2.0 open-source license, allowing commercial use and customization across a wide range of hardware.

Gemma 4 is available in four configurations: Effective 2B (E2B), Effective 4B (E4B), 26B Mixture of Experts (MoE), and 31B Dense. The larger models achieve top-tier performance for their size, with the 31B model ranking third among open models on the Arena AI text leaderboard. The E2B and E4B models are optimized for mobile and edge devices, supporting multimodal inputs and operating fully offline with low latency.

The new family supports features such as multi-step reasoning, native function calling, structured JSON output, and long-context processing up to 256K tokens. All models handle text, images, video, and audio, with built-in capabilities for optical character recognition and chart interpretation. They are trained on over 140 languages to support global applications.

Gemma 4 can be accessed through Google AI Studio and the Google AI Edge Gallery, with support from major AI frameworks including Hugging Face, vLLM, Ollama, and NVIDIA NIM. Developers can deploy the models locally or scale production on Google Cloud using Vertex AI and TPU acceleration. The release also includes collaborations with academic institutions and hardware partners such as Qualcomm, MediaTek, and NVIDIA to extend on-device AI performance.