# Gemini Nano Google's smallest Gemini language model variant, designed to run on-device on phones, laptops, and inside Chrome. The model behind Chrome's experimental [[Prompt API]] and [[Writing Assistance APIs]]. Demonstrates the [[Browser-Provided Language Models]] pattern in practice. ## Where It Runs | Platform | How | |---|---| | Pixel phones | Bundled with Android System Intelligence | | ChromeOS | OS-level component | | Chrome desktop | Downloaded on-demand via Chrome's component updater | | Chromium-based browsers (Edge, etc.) | Available where Chromium ships it | ## Capabilities - Text generation and summarization - Translation - Rewriting and proofreading - Multimodal (text + image) input on supported devices - Tool calling (function invocation) - Structured outputs Optimized via [[Knowledge Distillation]] from larger Gemini models, plus [[AI Quantization]] to fit within the memory and compute budget of consumer devices. ## Why It Matters - **Privacy**: prompts stay on-device; nothing sent to Google servers - **Latency**: no network round-trip - **Cost**: free for developers using Chrome's Prompt API - **Offline**: works without internet once downloaded ## Limitations - Smaller context window than cloud Gemini - Less capable than Gemini Pro/Ultra - Significant first-run download (~2-3 GB) - Requires capable hardware (NPU strongly preferred) ## Relationship to Web Standards Gemini Nano is the Google-specific model that powers Chrome's implementation of the W3C [[Prompt API]]. Other browsers may use different on-device models (Edge: integration with Windows Copilot runtime; Safari: likely Apple Intelligence's foundation model). ## References - https://deepmind.google/technologies/gemini/ - https://developer.chrome.com/docs/ai/built-in ## Related - [[Prompt API]] - [[Browser-Provided Language Models]] - [[Writing Assistance APIs]] - [[Large Language Models (LLMs)]] - [[Small Language Models (SLMs)]] - [[Knowledge Distillation]] - [[AI Quantization]] - [[On-Device Machine Learning]] - [[Apple Intelligence]] - [[Windows Copilot Runtime]] - [[Apple Neural Engine]] - [[Neural Processing Unit (NPU)]]