Chrome installs 4GB Gemini Nano model file for on-device AI features without clear user notice

Google Chrome is automatically downloading a 4GB model file to power its on-device AI features, catching users off guard with unexplained storage drops. The weights.bin file contains parameters for Google's Gemini Nano AI model, which powers Chrome features including scam detection, writing assistance, autofill, and suggestions.

Storage impact

The 4GB file is installed in Chrome's OptGuideOnDeviceModel directory when certain AI features are enabled. Users discovering the file report no clear notification about storage requirements at the point of enabling these features.

Because Gemini Nano runs locally rather than using cloud-based inference, it requires the full model weights stored on the user's device. This approach provides privacy benefits by keeping data processing on-device, but creates storage pressure particularly for users with limited disk space.

How to check and remove

Users can verify if the file exists by navigating to Chrome's data folders and checking the OptGuideOnDeviceModel directory for weights.bin. Simply deleting the file won't provide a permanent solution—Chrome will re-download it if AI features remain enabled.

To permanently remove the file and prevent re-download, users must navigate to Settings > System and disable the "On-Device AI" toggle. This removes all Gemini Nano-powered features from Chrome.

Disclosure concerns

Google does note that "Gemini Nano's exact size may vary as the browser updates the model" in its documentation for built-in AI features. However, this information appears in a lengthy feature guide rather than at the point where users enable AI features in settings.

Google has not yet commented on whether it will improve storage requirement disclosures or offer cloud-based alternatives for Chrome AI features. The Verge has reached out to Google for comment.

What this means

This situation highlights a growing tension in on-device AI deployment: local processing offers privacy advantages but requires significant storage that many users may not anticipate. As browser vendors add more AI capabilities, clear communication about system requirements becomes critical. For Chrome specifically, 4GB represents substantial overhead—roughly equivalent to 1-2 high-quality movies or thousands of documents. Users on devices with limited storage, particularly older laptops or Chromebooks, may find themselves forced to choose between AI features and available disk space.