Article complete

Google Now Lets You Use AI Without Internet on Smartphones

The Edge Gallery app is now available for Android, and an iOS variant is coming soon. 

A magnifying glass hovering over the Google logo, with search bar elements and icons in a colorful, abstract background.

Key takeaways

  • Download and use Google's Edge Gallery app for offline AI capabilities on smartphones.
  • Access multiple AI models, including Gemma 3 and Qwen 2.5, within the app.
  • Engage with AI for conversations, image understanding, and single-turn prompts without internet access.
  • Choose from lightweight and more capable models based on storage and performance needs.
  • Sign in to Hugging Face to download and utilize open-source AI models on your device.

Takeaways drafted with AI from the article text.

The Android app can be downloaded and installed through the Google AI Edge GitHub repository. The company announced that an iOS app will be available soon. 

“This app is a resource to understand the practical implementation of the LLM Inference API and the potential of on-device generative AI,” said the company in the blog post. 

AIM was able to download and install the application on an Android device. The app lets you download its Gemma 3 and 3n models, along with Alibaba’s Qwen 2.5 models. Users can install models across three categories: to have conversations, to upload and understand images, and to use ‘Prompt Lab’, a single-turn LLM in which users can ask one prompt, and the model provides one response. 

Models of varying sizes are available based on their capabilities; users can download a lightweight model of about 500 MB or a more capable model of approximately 4 GB. To download these models on the Edge Gallery app, users must sign in and agree to usage terms on Hugging Face, a platform where these open-source models are deployed.  

Once the model is installed, users can access and engage in conversations with the AI model without an internet connection. 

One of the models that Google provides in the Edge Gallery app is the recently announced Gemma 3n model. The small language model is built to operate locally on smartphones, using minimal memory and resources. 

Source: Google 

On the LMArena leaderboard for text-based tasks, the Gemma 3n scored 1293 points. In comparison, OpenAI’s o3-mini scored 1329 points, and the o4-mini scored 1379 points. The leading model is the Gemini 2.5 Pro model with 1446 points. 

Given that the model will not use the internet to fetch real-time information, it will only be able to provide information based on its knowledge cutoff date. For instance, the Gemma 3n model only contains information in its training data up to June 2024. 

Published

Spotted an error? Tell our editors. Read our editorial standards and corrections.