Google recently announced the release of Gemma 4 E2B for TPU, an advanced open model designed specifically to operate natively on the Tensor Processing Unit (TPU) found in the Pixel 10 series. The Gemma family represents Google’s collection of open models capable of functioning directly on user devices.
The announcement was made at I/O Connect India, following a recent satellite event held by Google in Berlin. Gemma 4 itself debuted in April, and Google had previously indicated that it would serve as the foundational model for the forthcoming Gemini Nano 4.
Technical Specifications and Performance
This new variant is built to run on the Tensor G5’s TPU. Google characterized the model as “state-of-the-art, powerful, yet remarkably lightweight.” The functionality is supported across several devices in the lineup: the Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, and Pixel 10 Pro Fold.
Enhanced On-Device Multimodal Capabilities
The optimized model unlocks several powerful multimodal features that operate entirely on the device, ensuring high levels of privacy. These capabilities include:
- AI Chat: Enabling deep, instantaneous conversations offline, even in extreme conditions like 30,000 feet.
- Ask Image: Allowing users to take a photo and identify objects, flora, or potential issues without requiring any internet connection (0% connectivity needed).
- Ask Audio: Providing 100% private, on-device transcription services for recording lectures or taking notes.
Furthermore, Google demonstrated “Mobile Actions,” a feature allowing users to issue commands—such as accessing Maps or controlling Wi-Fi—using only private voice input or text.
Real-World Application Use Cases
The integration of Gemma 4 E2B for TPU opens up several practical applications across various industries:
- Retail: Users can convert simple recipe ideas into detailed, localized shopping guides and maps that function completely offline while in a store.
- Automotive: The technology enables mechanics to perform immediate visual diagnostics simply by taking photographs of faulty components.
