Google today announced Gemma 4 E2B for TPU, a version of its Gemma family of open models optimized to run natively on the Pixel 10’s on-device Tensor Processing Unit (TPU). Gemma represents Google’s effort to provide powerful, privacy-focused AI that can operate directly on users’ devices, reducing dependence on cloud services and improving responsiveness for many everyday tasks.
The new Gemma 4 E2B variant was revealed during Google’s I/O Connect India event, following other recent product presentations at additional satellite events. This release extends the Gemma 4 lineup introduced earlier this year and highlights Google’s push to make advanced AI capabilities available locally on consumer hardware.
Gemma 4 serves as a foundation model in Google’s on-device AI strategy and forms the basis for more compact derivatives intended for mobile use. The E2B variant is tuned specifically for the Tensor G5 TPU found in the Pixel 10 series, and Google describes it as a state-of-the-art yet lightweight model built to deliver strong performance while remaining efficient enough for on-device execution.
The model is supported on the Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, and Pixel 10 Pro Fold. By running on the device’s TPU, Gemma 4 E2B enables a range of multimodal features without relying on an internet connection, which enhances privacy and ensures functionality in areas with limited connectivity.
Key multimodal capabilities made possible by this on-device model include:
- AI Chat: Fast, context-aware conversational experiences that work entirely offline, suitable for private use during travel or in restricted connectivity environments.
- Ask Image: The ability to analyze photos to identify objects, plants, signs, and other visual details without sending images to external servers, preserving user privacy.
- Ask Audio: On-device audio transcription and processing for lectures, meetings, or voice notes, keeping sensitive audio content local to the device.
Google demonstrated features such as Mobile Actions, which allow users to control core phone functions—like toggling Wi‑Fi or opening navigation—using private voice or text commands processed entirely on the device. These kinds of capabilities aim to streamline common tasks while minimizing data exposure to external services.
Beyond consumer convenience, on-device models open useful real-world applications across industries. For example:
- Retail: Shoppers can turn ideas or recipes into localized, offline shopping maps and product lists tailored to a specific store, supporting faster in-store navigation without sharing personal data.
- Automotive and repair: Technicians can receive instant visual diagnostics from photos of faulty parts, enabling quicker troubleshooting and guidance while keeping images and diagnostic data on the mechanic’s device.
Join 9to5Google Pro to get more from your favorite site
9to5Google continues to be a regular source for Android, Google, and broader tech coverage. For readers seeking extra content and community access, the publication offers an optional membership tier called 9to5Google Pro. Benefits of that membership include exclusive audio content, an ad-free members-only podcast, a private community, and direct Q&A sessions with editors.
Members gain access to a range of perks designed for enthusiasts who want deeper engagement with news and analysis. Subscribing also helps support continued reporting and product coverage across the site.
As on-device models like Gemma 4 E2B become more capable, we can expect a wider range of private, responsive, and offline-first experiences on modern smartphones. That shift has implications for user privacy, app responsiveness, and the kinds of applications developers can build when powerful AI runs directly on consumer hardware.