Google introduces Gemma 4, a new open AI model designed for Android devices, enhancing on-device AI capabilities. This model serves as the foundation for the upcoming Gemini Nano 4, promising seamless integration and future updates. Gemma 4 offers improved multilingual support, covering over 140 languages for a global reach. It boasts multimodal understanding, handling text, images, and audio within applications. Developers can choose between two sizes: E4B for complex tasks and E2B for speed. Gemma 4 is significantly faster and more battery-efficient than prior versions, with increased performance. Key improvements include enhanced reasoning, math skills, and time understanding. The new model also offers better image understanding, enhancing OCR capabilities. Developers can access Gemma 4 through the AICore Developer Preview to begin testing and building. They can also use Android Studio's ML Kit Prompt API to target either the E2B or E4B models. Google will release updates during the preview, including tool calling, structured outputs, and more. Testing is available on AICore-enabled devices with performance optimizations, and on other devices through the AI Edge Gallery app.