The Pixel 10’s on-device AI is now much more powerful, Gemma 4 optimized for Tensor G5

Google has announced the launch of Gemma 4 E2B, a specific variant of its open model optimized to run natively on the Pixel 10’s TPU.

This evolution promises to drastically increase on-device computational capabilities, delivering fast, secure processing and completely independent from an Internet connection.

Google introduces Gemma 4 E2B, optimized for the Pixel 10

During the I/O Connect India event, held not long after another major event in Berlin, Google unveiled a notable development for its hardware ecosystem.

Gemma 4 E2B represents the latest iteration of the company’s open-model family, configured to interact with the Tensor G5 chip architecture.

This version serves as the base for the future integration of Gemini Nano 4, introducing unprecedented efficiency in on-device processing.

This targeted optimization enables full leverage of the capabilities of the Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, and Pixel 10 Pro Fold, minimizing latency times and the impact on energy consumption.

What can Gemma 4 E2B do?

The integration of this architecture unlocks a range of practical applications that do not require any data to be transmitted to external servers, safeguarding the privacy of information.

Users can start in-depth textual conversations with the virtual assistant even in total isolation, for example during a flight.

The Ask Image feature enables you to frame objects, plants, or problematic elements to obtain immediate offline recognition.

The analysis also extends to audio, providing private and local transcription of lessons, interviews, and voice notes. Complementing this are the Mobile Actions, which allow you to manage certain system functions with simple voice or text commands, such as enabling Wi-Fi or viewing maps.