Tech companies like OpenAI, Anthropic, Meta, and Google are in a race to create the best large language model (LLM) using the training data at their disposal, with the goal to create useful services that capture the attention of consumers, businesses, and investors. The most powerful of these LLMs require computing power that just isn’t available on the tiny smartphones we carry in our pockets, which is why some companies offer smaller versions of their LLMs that can run locally on mobile devices.
The mobile optimized version of Google’s LLM is called Gemini Nano, and it’s currently available on a handful of Android phones. When Google first released it in December, Gemini Nano was only available on a single smartphone, the Pixel 8 Pro. It rolled out a month later to Samsung’s Galaxy S24 series, ending the brief Pixel exclusivity on the AI model. Throughout 2024, Google rolled out the Gemini Nano model to a handful of additional Pixel and Samsung devices. Last week, though, Google finally brought the AI model to a smartphone from another manufacturer: Xiaomi.
It seems the Xiaomi 14T was just the beginning, though, as Google has opened the floodgates and started rolling out Gemini Nano support to a variety of new phones. This includes phones from Nano newcomers like Motorola and Realme as well as phones from existing Nano supporters like Samsung.
Here’s the list of devices that currently support Gemini Nano, as of October 1, 2024.
What phones support Gemini Nano?
Devices that support Gemini Nano 1
- Google Pixel 8
- Google Pixel 8 Pro
- Google Pixel 8a
- Motorola Edge 50 Ultra
- Motorola Razr 50 Ultra
- Samsung Galaxy S24
- Samsung Galaxy S24+
- Samsung Galaxy S24 Ultra
- Samsung Galaxy S24 FE
- Samsung Galaxy Z Flip 6
- Samsung Galaxy Z Fold 6
- Realme GT 6
- Xiaomi 14T
- Xiaomi 14T Pro
- Xiaomi MIX Flip
Devices that support Gemini Nano 2
- Google Pixel 9
- Google Pixel 9 Pro
- Google Pixel 9 Pro XL
- Google Pixel 9 Pro Fold
What’s the difference between Gemini Nano 1 and 2?
Although both Gemini Nano 1 and Gemini Nano 2 are part of the Gemini 1.0 family of LLMs, Gemini Nano 2, also known as Gemini Nano with multimodality, offers support for image, speech, and audio input in addition to text. According to Google, Nano 2 is about twice the size of Nano 1 and delivers much better performance in academic benchmarks and real-world applications.
Performance of Gemini Nano 1 vs. Gemini Nano 2. Source: Google.
Gemini Nano 2 is currently only available on the Google Pixel 9 series, with no word on when or if it’ll come to other devices.
What AI features are powered by Gemini Nano?
Gemini Nano 2 on the Google Pixel 9 series powers AI features in the Pixel Screenshots, Talkback, Pixel Recorder, Phone by Google, and Pixel Weather apps. Pixel Screenshots uses Gemini Nano 2 to analyze your screenshots locally. TalkBack uses it to generate descriptions of images. Pixel Recorder uses it to summarize audio recording transcripts. Phone by Google uses it to summarize phone calls. Lastly, Pixel Weather uses it to generate a summary of the day’s weather.
Gemini Nano 1 on the Google Pixel 8 series is a bit more limited due to its smaller size and text-only input support. It helps the Gboard app generate smart replies in supported messaging apps. (Gemini 2 on the Google Pixel 9 also powers smart replies in Gboard with no noticeable differences.) It also helps the Pixel Recorder app to generate summaries of audio recording transcripts, though it can’t handle audio recordings that are longer than about 15 minutes.
On non-Pixel devices, the Gemini Nano 1 model only powers a single AI feature right now: Magic Compose in Google Messages. This feature takes your last 20 messages in a conversation as context and generates some suggested replies you can send to a contact. Because this happens on-device, your texts never leave your device.
Although Magic Compose is currently the only Gemini Nano-powered feature that’s common across all devices that support the AI model, that hopefully won’t be true for long. In fact, Google today just opened up experimental access to the AI Edge SDK that lets apps use Gemini Nano. However, this experimental access is only “for development purposes,” so third-party app developers aren’t yet able to deploy AI features that make use of Gemini Nano. Plus, this experimental access only gives developers access to text-to-text prompting via Gemini Nano 2, so most devices are left out of this experiment.
Regardless, it’s clear that Google is serious about bringing Gemini Nano to more Android phones. As of today, there are now 15 Android phones that support Gemini Nano 1 and 4 that support Gemini Nano 2. Google deploying the AI model to more phones will hopefully encourage developers to implement new and exciting features that take advantage of it.