Google unveils Gemma 4 AI model

Google DeepMind has introduced its latest artificial intelligence model, Gemma 4, marking a significant step forward in open AI development. The new release is being described as Google’s most advanced open model to date, designed specifically for complex reasoning tasks and agent-based workflows under a permissive Apache 2.0 license.

The company has launched Gemma 4 in four different sizes to cater to a wide range of use cases. These include effective 2B (E2B), effective 4B (E4B), a 26B Mixture of Experts (MoE), and a 31B Dense model. According to rankings on the Arena AI leaderboard, the 31B model currently stands among the top-performing open AI models globally, while the 26B variant has also shown strong performance despite its relatively smaller size.

Google highlighted that Gemma 4 offers improved efficiency, delivering high intelligence per parameter. Since the original Gemma models were released, they have gained massive traction, with over 400 million downloads and more than 100,000 community-built variants forming what the company calls the “Gemmaverse.”

Google Introduces Easy Switching Tool to Move ChatGPT Data to Gemini

One of the key features of Gemma 4 is its support for advanced functionalities such as native function calling, structured JSON outputs, and system-level commands. These capabilities allow developers to build intelligent agents that can interact with tools, APIs, and external systems more effectively.

The models are also multimodal, supporting text, image, and video processing, while smaller versions additionally enable audio input for speech recognition. With support for over 140 languages and extended context windows of up to 256K tokens, Gemma 4 allows developers to handle large datasets and long documents efficiently.

Optimized for mobile and edge devices, smaller models can run offline with minimal latency. Developers can access Gemma 4 through platforms like Hugging Face, Kaggle, Ollama, and Google AI Studio.