The Google AI Edge Gallery application, which allows open-source large language models to run directly on smartphones, has been updated with highly anticipated Gemma 4 support. Available on both iOS and Android, this app continues to push the boundaries of on-device AI. Recent speed tests have highlighted massive performance gaps between today’s flagship smartphones.
Google AI Edge Gallery Empowered by Gemma 4
According to Google’s official announcement, this update brings full and official support for the Gemma 4 family. The company emphasizes that users can experience advanced reasoning, logic, and creative capabilities without sending any data to a remote server. Google DeepMind CEO Demis Hassabis noted that Gemma 4 brings together the best open models in its class, offering four customizable options:
- 31B: For high-performance tasks.
- 26B MoE (Mixture of Experts): For faster responses.
- 2B and 4B: Ideal for mobile devices and lightweight tasks.
The Flagship Showdown: Gemma 4 Speed Test
Following the integration of Gemma 4, tech platform TechDroider conducted a comprehensive on-device AI speed test. The contenders included the iPhone 17 Pro Max, Samsung S26 Ultra, Pixel 10 Pro XL, OnePlus 15, and Xiaomi 17 Pro Max. The results were quite startling.
Task 1: HTML Generation
In the coding task, the OnePlus 15 took the lead, completing the process in just 2.6 minutes. It was followed by:
- iPhone 17 Pro Max: 3.1 minutes
- Xiaomi 17 Pro Max: 3.4 minutes
- Samsung S26 Ultra: ~4 minutes
- Pixel 10 Pro XL: 15+ minutes (A major disappointment for Google’s own hardware).
Task 2: Visual Processing (Ask Image Model)
When switching to the Gemma 4 “Ask Image” task, the leaderboard shifted significantly. The iPhone 17 Pro Max dominated the field, finishing the task in a mere 10 seconds. The rest of the pack followed:
- Samsung S26 Ultra: 14.2 seconds
- OnePlus 15: 14.4 seconds
- Xiaomi 17 Pro Max: 17 seconds
- Pixel 10 Pro XL: 59.9 seconds
These results prove that raw processing power and deep optimization are vital for running local AI models. As an academic interested in technology and communication, do you think the Pixel 10 Pro XL’s lag is a sign of hardware limitations, or is Google perhaps prioritizing “quality over speed” in its AI processing? Share your thoughts in the comments!
Since you follow smartphone trends and use an RTX 3070 PC, would you like me to research how the mobile Gemma 4 4B model compares to the PC-based Gemma 4 31B in terms of logic accuracy? Let me know!
