Gemini Nano

Google’s lightweight on-device AI model, used as the foundation for Chrome’s Prompt API.

Specifications

• GPU version: ~4GB (HIGHEST_QUALITY or FASTEST_INFERENCE modes)
• CPU version: ~2.7GB
• Deployment: Silently installed in Chrome profile directory (OptGuideOnDeviceModel/weights.bin) since 2024
• Purpose: Fraud detection, developer API support, local inference without cloud dependency

Controversy

The model’s silent installation on hundreds of millions of devices without consent sparked significant privacy debate in May 2026. Key issues:

• No user consent or notification
• Auto-restore on deletion
• Environmental impact: ~0.06 kg CO₂e per device transmission
• Google claims opt-out option added in February 2026, but core transparency issues remain unresolved