Google Tensor SDK Beta with LiteRT Unlocks On-Device ML for Pixel 10

The Google Tensor ML SDK, designed to harness the power of Google Pixel's custom-designed Tensor System-on-Chip (SoC) and its dedicated Tensor Processing Unit (TPU), has officially moved from Experimental Access Program (EAP) to Beta. This significant milestone, announced on May 19, 2026, brings a new era of on-device machine learning capabilities to developers targeting the Google Pixel 10 family of devices and beyond. A key aspect of this Beta launch is the integration with LiteRT, Google's cutting-edge on-device framework for high-performance ML deployment on edge platforms. LiteRT simplifies the developer experience by abstracting complex, vendor-specific SDKs, compilers, and runtimes into a unified, streamlined API. This means developers can now convert, compile, deploy, and run PyTorch or TFLite models optimized for the Tensor TPU with unprecedented ease. The Beta launch offers two major advantages: a unified developer workflow with LiteRT and access to a Model Garden featuring over 100 models optimally configured for Tensor's TPU. These advancements unlock a range of interactive, real-time, and private on-device AI experiences, including enhanced camera features like Pro Zoom, intelligent communication tools such as Voice Translate, and productivity aids like Call Notes. Developers can also build features for intelligent content creation, vision and understanding, and audio and accessibility, pushing the boundaries of what's possible with on-device AI.