May 29, 2025 — Google is reshaping how we interact with artificial intelligence by unveiling Edge Gallery, a new app that allows users to run AI models directly on their smartphones—completely offline. Currently available for Android devices via the Google AI Edge GitHub repository, the iOS version is expected to be released in the near future.
This move marks a significant shift toward on-device generative AI, offering users more control, privacy, and performance efficiency without relying on an internet connection.
May 29, 2025 — Google is reshaping how we interact with artificial intelligence by unveiling Edge Gallery, a new app that allows users to run AI models directly on their smartphones—completely offline. Currently available for Android devices via the Google AI Edge GitHub repository, the iOS version is expected to be released in the near future.
This move marks a significant shift toward on-device generative AI, offering users more control, privacy, and performance efficiency without relying on an internet connection.
AI On-the-Go: No Internet Required
Edge Gallery is designed to demonstrate the capabilities of Google’s LLM Inference API and brings a hands-on, functional approach to using large language models (LLMs) on mobile devices. Once installed, the app enables users to download and run AI models locally, eliminating the need for constant data access or cloud-based computation.
“Edge Gallery is a resource to understand the practical implementation of the LLM Inference API and the potential of on-device generative AI,” Google noted in its official blog post.
Edge Gallery is designed to demonstrate the capabilities of Google’s LLM Inference API and brings a hands-on, functional approach to using large language models (LLMs) on mobile devices. Once installed, the app enables users to download and run AI models locally, eliminating the need for constant data access or cloud-based computation.
“Edge Gallery is a resource to understand the practical implementation of the LLM Inference API and the potential of on-device generative AI,” Google noted in its official blog post.
What’s Inside the Edge Gallery App?
Upon installing the app, users gain access to a suite of models, including:
- Google’s Gemma 3 and Gemma 3n
- Alibaba’s Qwen 2.5
These models can be downloaded across three primary use cases:
- Conversational AIfor chatting and virtual assistant-style interactions
- Visual analysisfor uploading and interpreting images
- Prompt Lab, which enables one-turn text generation where the model responds to a single query
Upon installing the app, users gain access to a suite of models, including:
- Google’s Gemma 3 and Gemma 3n
- Alibaba’s Qwen 2.5
These models can be downloaded across three primary use cases:
- Conversational AIfor chatting and virtual assistant-style interactions
- Visual analysisfor uploading and interpreting images
- Prompt Lab, which enables one-turn text generation where the model responds to a single query
Lightweight to Powerful: Choose the Right Model
Edge Gallery offers flexibility in terms of model sizes and performance:
- Lightweight versions(~500 MB) for basic tasks and minimal resource use
- More advanced models(~4 GB) for richer and more complex interactions
To access these models, users are required to log in and agree to terms on Hugging Face, the popular platform hosting these open-source AI models.
Once installed, these AI models function entirely offline. This makes the Edge Gallery app particularly appealing to users concerned about data privacy, limited connectivity, or those in regions with poor internet infrastructure.
Edge Gallery offers flexibility in terms of model sizes and performance:
- Lightweight versions(~500 MB) for basic tasks and minimal resource use
- More advanced models(~4 GB) for richer and more complex interactions
To access these models, users are required to log in and agree to terms on Hugging Face, the popular platform hosting these open-source AI models.
Once installed, these AI models function entirely offline. This makes the Edge Gallery app particularly appealing to users concerned about data privacy, limited connectivity, or those in regions with poor internet infrastructure.
Spotlight on Gemma 3n: Small but Smart
Among the available models, Gemma 3n stands out as a compact, efficient AI solution optimized for mobile hardware. Despite its small footprint, it performs competitively in benchmarking tests. On the LMArena leaderboard—a widely recognized benchmark for text-based AI tasks—Gemma 3n scored 1293 points, placing it just behind:
- OpenAI’s o3-mini: 1329 points
- OpenAI’s o4-mini: 1379 points
- Gemini 2.5 Pro(leading model): 1446 points
While it doesn’t rely on internet connectivity, it’s important to note that these models operate within the constraints of their knowledge cutoffs. For instance, Gemma 3n has been trained only on data available up to June 2024, meaning it cannot provide real-time updates or access current events beyond that date.
Among the available models, Gemma 3n stands out as a compact, efficient AI solution optimized for mobile hardware. Despite its small footprint, it performs competitively in benchmarking tests. On the LMArena leaderboard—a widely recognized benchmark for text-based AI tasks—Gemma 3n scored 1293 points, placing it just behind:
- OpenAI’s o3-mini: 1329 points
- OpenAI’s o4-mini: 1379 points
- Gemini 2.5 Pro(leading model): 1446 points
While it doesn’t rely on internet connectivity, it’s important to note that these models operate within the constraints of their knowledge cutoffs. For instance, Gemma 3n has been trained only on data available up to June 2024, meaning it cannot provide real-time updates or access current events beyond that date.
A Step Toward Decentralized AI
The introduction of Edge Gallery signifies Google’s commitment to pushing the boundaries of AI democratization and decentralization. By making powerful AI tools accessible without internet access, Google is enabling a new era of offline productivity, privacy-focused interactions, and AI-driven creativity—even in the most remote locations.
With the iOS version around the corner, Edge Gallery is poised to become a cornerstone in how users engage with AI—anytime, anywhere, without the cloud.
Read more: India’s Top 10 Women Leaders in Global Capability Centres (GCCs) – 2025 Edition
The introduction of Edge Gallery signifies Google’s commitment to pushing the boundaries of AI democratization and decentralization. By making powerful AI tools accessible without internet access, Google is enabling a new era of offline productivity, privacy-focused interactions, and AI-driven creativity—even in the most remote locations.
With the iOS version around the corner, Edge Gallery is poised to become a cornerstone in how users engage with AI—anytime, anywhere, without the cloud.
Read more: India’s Top 10 Women Leaders in Global Capability Centres (GCCs) – 2025 Edition