Gemini Nano

Definition

Gemini Nano is the smallest, on-device member of Google's Gemini model family, designed to run efficiently on smartphones and other edge hardware for tasks like summarization, smart reply, and text proofreading without requiring a network connection.

Overview

Announced alongside the broader Gemini family in December 2023, Gemini Nano is distilled from larger Gemini models to fit within the memory and compute constraints of mobile devices. It ships in two sizes, Nano-1 (1.8 billion parameters) and Nano-2 (3.25 billion parameters), and is optimized to run through Google's AICore system service on Android, which manages model execution, updates, and hardware acceleration (via a device's neural processing unit) so individual apps don't need to bundle their own model weights. Gemini Nano first shipped on the Pixel 8 Pro in late 2023, powering features like Summarize in Recorder, Smart Reply in Gboard, and Magic Compose in Google Messages, and has since expanded to additional Pixel and select Samsung Galaxy devices. Because inference happens entirely on-device, Gemini Nano offers lower latency, offline availability, and stronger privacy guarantees than cloud-based models for the tasks it supports, at the cost of being far less capable than Gemini Pro or Gemini Ultra on complex reasoning tasks. Google also exposes Gemini Nano to web developers through Chrome's built-in AI APIs (part of the Chrome origin trial / built-in AI initiative), letting websites perform tasks like summarization or translation using the on-device model without sending data to a server. This positions Gemini Nano as a direct competitor to Apple's on-device foundation models used in Apple Intelligence, and to other small edge-optimized models such as Microsoft's Phi series, in the broader industry push toward efficient, privacy-preserving on-device AI.

Key Features

• Two sizes: Nano-1 (1.8B parameters) and Nano-2 (3.25B parameters)
• Runs fully on-device via Android's AICore system service
• Distilled from larger Gemini models for mobile-scale efficiency
• Enables offline AI features with lower latency than cloud inference
• Powers Pixel features like Summarize, Smart Reply, and Magic Compose
• Exposed to web developers through Chrome's built-in AI APIs
• Stronger on-device privacy guarantees since data need not leave the device

Use Cases

Alternatives

Frequently Asked Questions

Related in AI Models