Tensor G6 on Pixel 11: Significant TPU upgrade for AI running right on the device

In addition to hardware changes, Tensor G6 on the Pixel 11 line has been significantly upgraded by Google in its Tensor Processing Unit (TPU) processor. According to the company, the new TPU has 50% more computing power and can be up to 3.5 times faster in some tasks, aiming for AI processing right on the device instead of relying entirely on the cloud. In the Tensor system, the CPU handles applications, the GPU mainly handles games, and the TPU is designed specifically for machine learning tasks with the ability to calculate matrices and high-speed memory access. This is the main AI processing block on Pixel, serving local operating features such as voice recognition, speech-to-text conversion and input suggestions. TPU is also deeply integrated into Tensor's image processing chain. Google uses it for computational photography tools like Night Sight, Super Res Zoom, subject separation in portraits, and video processing. This upgrade is one of the factors that helps Pixel 11 expand the zoom level to 120x, compared to 100x on the previous generation. On Pixel 11, Gemini Nano is said to be able to respond faster and give better results in small queries. The phone also adds a voice-to-text feature that can automatically remove filler words like “um,” translate sign language to text in the Camera app, as well as Magic Capture: take up to 500 photos and then use Gemini to choose six suggestions of five photos and one video. Live Translate can also translate videos, podcasts and voice messages into the language the user chooses while watching or listening. Another notable point is Proactive Assistance, a tool that can proactively display contextual information from certain applications such as contacts or messages to support schedule management and respond to emails and messages faster. Google says the processing takes place on the device, making it possible for data to be kept private and encrypted. As of mid-2026, third-party programmers will be able to start exploiting TPU on Tensor chips through LiteRT and the Google Tensor SDK beta. A 50% increase in computing power does not mean every AI application is exactly 50% faster, but it will create more room to run larger models or heavier tasks right on the phone, such as translation, transcription, object recognition, image processing and generative AI. However, not all AI features on the Pixel have moved to local processing. Gemini Intelligence still needs to use information and processing from the cloud in some cases, the “Banana” image generator and Video Boost also continue to be cloud-dependent features. The new TPU therefore cannot replace the capacity of the server infrastructure, but is a step forward for Pixel 11 to handle more AI tasks with lower latency, the ability to operate without a network and better privacy advantages. Android Authority.