Google is introducing its Gemini AI model to the Pixel 8 Pro, marking a notable step in bringing advanced language models to mobile devices. Gemini, considered one of the most powerful large language models (LLM) developed by Google, was initially designed for data centers due to its size and complexity.
However, with the introduction of Gemini Nano, a scaled-down version, the model can now operate locally and offline on devices, starting with the Pixel 8 Pro. Google envisions Gemini Nano becoming an integral part of the Android platform in the future.
For Pixel 8 Pro users, the December Feature Drop includes two applications powered by Gemini Nano: the auto-summarization feature in the Recorder app and the Smart Reply function in the Gboard keyboard. These features operate offline on the device, ensuring speed and a native feel.
Despite its potential, Gemini Nano’s initial rollout is limited — The Smart Reply feature is currently Gemini-powered only in WhatsApp, with plans to expand to more apps next year; Additionally, Gemini is currently available only in English, limiting accessibility for non-English users.
Gemini Nano, being the smallest version of the Gemini models, aims to maintain high capability within the constraints of a mobile device’s footprint, addressing concerns related to memory and speed limitations. Google’s Tensor 3 processor is currently the sole processor capable of running the model, but the company is working on integrating Nano into Android through a new system service called AICore.
This service enables developers to incorporate Gemini-powered features into their apps, with companies like Qualcomm, Samsung, and MediaTek producing compatible processors.
While Google has emphasized the AI capabilities of its Pixel phones in recent years, the introduction of Gemini Nano suggests a broader application for high-end Android devices in the future. Although currently exclusive to the Pixel 8 Pro, the expansion of Gemini Nano to more devices appears to be on the horizon, potentially transforming them into AI-enhanced devices over time.