Gemma 3n is a significant step forward from Google's previous AI models. After being introduced at I/O last month, the full version is now available and ready to run directly on mobile devices.

Gemma is an open-source AI model kit from Google. It differs from Gemini in that it allows developers to download and edit it.

w2q61pab.png
On-device AI processing inherently requires large amounts of RAM. Photo: Shutterstock

The US tech company said Gemma 3n can process inputs such as images, audio and video to generate text in place, a leap forward from the text-only approach as before. Another highlight is that it runs on hardware with modest memory, requiring only 2GB, and is better at handling tasks such as programming and reasoning.

Gemma 3n supports 140 languages ​​for text and 35 languages ​​for multimodality.

Google is using a new architecture called MatFormer to increase efficiency. It’s like a Russian Matrioska doll: a larger model contains smaller, fully functional versions inside. This allows for scaling (E2B, E4B) for different use cases.

In terms of performance, Google said the E4B model scored 1,300 LMArena points, making it the first model under 10 billion parameters to reach this milestone.

The model's audio functionality now supports speech-to-text conversion and on-device translation, using an encoder that can process speech in detail. Visually, the MobileNet-V5 encoder can process video at up to 60FPS on Pixel devices.

At Google I/O, Google announced that Gemma 3n will run efficiently on everyday devices like smartphones, laptops, and tablets. With this breakthrough, phones with 2GB to 3GB of RAM can also run AI, instead of having to compete for RAM, bringing AI to a wider audience.

Multiple reports suggest that manufacturers are ramping up RAM on mobile devices to meet AI demands. According to Macquarie Research, RAM capacity on Android flagships is on the rise. The company predicts that the Galaxy S26 series will increase to 16GB of RAM. Meanwhile, the iPhone also comes with 8GB of RAM to run Apple Intelligence and is expected to increase to 12GB of RAM next year.

(According to Neowin)

Touch the back of your iPhone to open any app, no need to touch the screen. The Touch Back feature of your iPhone helps users open apps, control functions or perform quick tasks without touching the screen.

Source: https://vietnamnet.vn/dien-thoai-2gb-cung-co-the-dung-ai-voi-dot-pha-moi-cua-google-2415679.html