Google announced Gemma 4 E2B optimized for the Pixel 10's TPU, enabling on-device multimodal AI capabilities like offline chat, image recognition, and audio transcription.
Gemma 4 E2B for TPU is designed to run natively on the Pixel 10's TPU. It supports instant, deep offline conversation, image identification, and on-device audio transcription. The model also lets users command core phone functions, like Wi-Fi or maps, using just private voice or text. Video demonstrations of what the model can do are available in the article.
# Google announces Gemma 4 optimized for the Pixel 10’s TPU
Source: [https://9to5google.com/2026/07/14/pixel-10-gemma-4/](https://9to5google.com/2026/07/14/pixel-10-gemma-4/)
Google today[announced](https://developers.googleblog.com/unlocking-the-next-era-of-on-device-ai-with-google-tensor-and-pixel/)[Gemma 4 E2B for TPU](https://huggingface.co/litert-community/gemma-4-E2B-it-litert-lm/blob/main/gemma-4-E2B-it_Google_Tensor_G5.litertlm)that’s designed to run natively on the Pixel 10’s TPU\. Gemma is Google’s series of open models that can run on\-device\.
This new model was announced at I/O Connect India, with Google last week hosting another satellite event in Berlin\.
Gemma 4 was first[introduced in April](https://9to5google.com/2026/04/02/google-gemma-4/), with Google previewing that it is the base model for the upcoming[Gemini Nano 4 model](https://9to5google.com/2026/04/02/gemini-nano-4-android/)\.
The new variant released today runs on the Tensor G5’s Tensor Processing Unit \(TPU\), with Google describing it as a “state\-of\-the\-art, powerful, yet remarkably lightweight model\.” It’s supported on the Pixel 10, 10 Pro, 10 Pro XL, and 10 Pro Fold\.
Multimodal capabilities make possible:
- **AI Chat**: Instant, deep offline conversations \(even at 30,000 feet\)\.
- **Ask Image**: Snap and identify objects, plants, or issues with 0% internet\.
- **Ask Audio**: 100% private, on\-device audio transcription for lectures and notes\.
Google demoed capabilities like Mobile Actions that let you “command core phone functions—like WiFi or maps—using just private voice or text\.”
In terms of real\-world applications:
- **Retail**: Converting recipe ideas into precise, localized in\-store shopping maps completely offline\.
- **Automobile**: Giving mechanics on\-the\-spot visual diagnostics from photos of faulty parts\.
---
### Join 9to5Google Pro to get more from your favorite site
We’re already your daily source for Android, Google, and tech news\. Now you can get more with**[9to5Google Pro](https://9to5google.com/2026/07/06/introducing-9to5google-pro/)**:
- *Pixelated Pro,*our new members\-only podcast \(ad\-free\)
- Members\-only Discord community
- Monthly Q&A with the editors
And there’s lots more to come\. Subscribe now for[$5/month](https://bit.ly/4velxXh)or[$50/year](https://bit.ly/44iEFsd)\!
---
[](https://google.com/preferences/source?q=https://9to5google.com)
*FTC: We use income earning auto affiliate links\.*[More\.](https://9to5mac.com/about/#affiliate)
Google introduces Gemma 3 270M, a compact 270-million parameter model designed for efficient on-device AI with strong instruction-following capabilities and extreme energy efficiency (0.75% battery for 25 conversations on Pixel 9 Pro).
Google AI Edge Gallery v1.0.13 & v1.0.14 updates add support for Gemma 4 with multi-token prediction, Pixel TPU optimization, experimental MCP, new skills, and chat history saving, enhancing on-device generative AI capabilities.
Google releases Gemma 4 12B, a compact AI model optimized for local laptop use with only 16GB of RAM, featuring multi-token prediction and streamlined multimodal capabilities for text, audio, and images.
Google introduces Gemma 3, a collection of lightweight open models (1B, 4B, 12B, 27B) designed to run on single GPUs or TPUs, featuring support for 140+ languages, 128k context window, and multimodal capabilities. The models outperform larger competitors like Llama 3 and DeepSeek-V3 while maintaining efficiency for on-device deployment.