Google has officially launched Gemma 4, a new open AI model family that aims to bring powerful artificial intelligence capabilities directly to everyday devices, including Android smartphones. Announced by Google DeepMind CEO Demis Hassabis on X, Gemma 4 is being positioned as one of the best open models available in its size category.
Unlike proprietary systems such as Gemini or OpenAI's latest models, Gemma 4 is fully open source, allowing developers to download, modify, and run the model locally. This marks a major shift toward accessible AI, as users can now operate advanced models without relying on cloud services or constant internet connectivity.
One of the biggest advantages of Gemma 4 is privacy. Since the model can run locally on devices, user data does not need to be shared with Google or any third party, addressing growing concerns around data security in AI systems.
What is Gemma 4?
Gemma 4 is a family of four purpose-built AI models designed for advanced reasoning and agent-based workflows:
E2B (2 Billion parameters): Optimised for lightweight tasks on mobile and edge devices
E4B (4 Billion parameters): Enhanced performance for smartphones and compact hardware
26B MoE (Mixture of Experts): Focused on low latency and efficiency
31B Dense Model: High-performance model ranked third globally among open models on the Arena AI leaderboard
Notably, the 31B model outperforms some models nearly 20 times its size, showcasing the efficiency improvements Google has achieved.
Advanced Capabilities
Gemma 4 introduces significant upgrades in AI performance, including:
Multi-step reasoning and complex logic handling
Support for agentic workflows, allowing users to delegate tasks to AI agents
Ability to create multiple AI agents that can collaborate on tasks
Strong offline code generation, turning devices into local AI coding assistants
The models also support multimodal inputs, including images and video, while smaller versions can process audio for speech recognition.
Built for Smartphones and Beyond
A standout feature of Gemma 4 is its ability to run across a wide range of hardware, from high-end GPU systems to regular Android smartphones. This is a foundational step toward bringing AI to billions of devices worldwide.
Gemma 4 is also expected to serve as the base for the next generation of Gemini Nano, Google's on-device AI model for Android, enabling offline AI experiences without internet dependency.
Open Source and Developer-Friendly
Google has released Gemma 4 under the Apache 2.0 license, offering developers complete flexibility to use, modify, and deploy the models. This is a significant change from earlier versions, which were limited to open-weight access under stricter terms.
Since its initial release, the Gemma ecosystem has already seen over 400 million downloads and more than 100,000 community-built variants, highlighting its rapid adoption among developers.
Gemma 4 supports over 140 languages and offers extended context windows, up to 128,000 tokens for smaller models and 256,000 tokens for larger ones, enabling it to handle long documents and complex codebases in a single prompt.
Availability
Developers can access Gemma 4 via Google AI Studio or download it from platforms such as Hugging Face, Kaggle, and Ollama.
With its open nature, strong performance, and ability to run locally, Gemma 4 could redefine how AI is used, shifting from cloud-dependent systems to powerful, private, on-device intelligence.
