Google Chrome Secretly Downloads Huge Local AI Models

Google Chrome automatically installs local neural network components on user systems via default configurations. The browser downloads a 4GB local version of Gemini Nano without explicit user consent. The system activates this model during interactions with Google services, and Google enables these options by default.

An X user, @zephyrianna, identified this behavior:

Is this on your PC too??

Finally found what kept filling up my C drive...a 4 GB "weights.bin" file. Looked it up, it's Google's Gemini Nano LLM that runs built-in AI APIs on your PC 😭

If you want it gone, after deleting the file, go to chrome://flags/ to disable these! pic.twitter.com/JMRBSZfwHW

— Zephyrianna 🐦💫 (@zephyrianna) February 20, 2026

This integration solely affects standard Chrome builds. Alternative browsers do not implement Gemini Nano, although they might utilize other AI models. These large model files appear within the C:Users<username>AppDataLocalGoogleChromeUser DataOptGuideOnDeviceModel directory.

Procedures for disabling the AI model in Chrome

• Close the Chrome application completely before altering any files or settings.
• Access the internal configuration settings by entering chrome://flags/ into the address bar and pressing Enter.
• Locate the setting Enables Optimization Guide On Device and set the configuration to Disabled.
• Locate the Prompt API setting and disable this option as well.
• Once these flags are disabled, delete the weights.bin file, which is approximately 4 GB, from the C:Users<username>AppDataLocalGoogleChromeUser DataOptGuideOnDeviceModel directory.

Users should exercise caution during this process. Incorrectly modifying browser directories may adversely affect application stability. Users should ensure they understand the function of any file before deletion. Instead of deleting the file immediately, users may rename it to verify system behavior after relaunching the browser.

The availability of this local model feature remains limited based on country, market, and language. Google is conducting a gradual roll-out, meaning the feature may not yet exist on all systems.

Recent iterations of Chrome utilize the Optimization Guide tool for on-device AI testing. Running the model locally may improve response times, reduce dependency on cloud services, and potentially improve privacy for specific tasks.