Google announced Gemma 4 E2B for TPU on July 14, 2026 at I/O Connect India, a variant of its open Gemma model family built to run natively on the Tensor G5's Tensor Processing Unit. It is supported on the Pixel 10, 10 Pro, 10 Pro XL and 10 Pro Fold, and weights are published on Hugging Face under the litert-community organization in LiteRT-LM format. Google described the model as lightweight and multimodal, citing offline chat, image identification and on-device audio transcription, and demoed Mobile Actions for controlling phone functions such as WiFi and maps by voice or text. Gemma 4 was first introduced in April 2026 and is the base model for the forthcoming Gemini Nano 4.