What Chrome’s Hidden 4GB Gemini Nano Model Is and Why It Matters
Many Chrome users are discovering a mysterious 4GB file called weights.bin buried deep in their browser folders. This file is the local Gemini Nano AI model that Chrome downloads to power on-device tools such as scam detection, autofill, writing assistance, and other AI helpers. Instead of sending your data to cloud servers, Gemini Nano runs locally and relies on those stored parameters. Technically, that can be better for privacy, but there’s a catch: for many users, this model arrived without a clear prompt, explicit opt-in, or obvious disclosure about its 4GB footprint. Chrome has been capable of fetching this model since 2024, and whether it installs depends on your hardware, account features, and the sites you visit. If you never asked for AI features, that silent download can feel like an unwelcome surprise—and a prime target for Chrome storage cleanup.
How to Check If Chrome’s Gemini Nano Model Is Taking Up Your Storage
Before you think about Gemini Nano removal, you need to confirm whether the model is actually on your system. The file typically lives in a folder named OptGuideOnDeviceModel inside your Chrome data directory and is called weights.bin. On many setups, you can locate it by browsing to your user’s Chrome profile folder and searching for “OptGuideOnDeviceModel” or for large files over 1GB. Some users have used system tools or terminal commands to scan their Chrome support directory and flag very large files. If you do not see weights.bin or the OptGuideOnDeviceModel folder, Chrome may not have installed the local AI model yet—especially if you disabled AI-related options early. If you do find the file, you are looking at roughly 4GB of reclaimable disk space and a clear candidate for 4GB storage recovery.
Turn Off Chrome’s On-Device AI So the Model Stops Coming Back
Simply deleting weights.bin is not enough. With Chrome’s on-device AI features enabled, the browser can quietly download the Gemini Nano model again the next time it thinks it is needed. To make your AI model delete permanent, you must first switch off local AI in Chrome settings. Google notes a dedicated toggle in Chrome’s System area that disables the on-device model entirely, removes it, and blocks future downloads. Another documented route is via Chrome’s experimental flags: type chrome://flags into the address bar, search for an entry named optimization-guide-on-device-model, and set it to Disabled. After restarting Chrome, the browser should unregister the local model. This combination of Chrome AI settings changes is essential if you want to prevent silent reinstalls and ensure your Gemini Nano removal actually sticks.
Manually Remove the 4GB Gemini Nano File to Reclaim Disk Space
Once the on-device AI switch is off, you can safely perform a Chrome storage cleanup by deleting the Gemini Nano file itself. Navigate to your Chrome profile or application support folder and find the OptGuideOnDeviceModel directory. Inside, you should see the weights.bin file, which is the roughly 4GB local AI model. Deleting this file will immediately free that space, and with the feature disabled, Chrome should not redownload it. Remember that a standard Chrome installation already consumes multiple gigabytes for cache, profile data, and extensions, so consider also pruning old profiles or unused add-ons while you are there. The key difference is that weights.bin exists purely for on-device AI; if you are not using Gemini-based features, removing it is a straightforward way to regain control over your storage without affecting basic browsing.
What Changed in 2026—and How to Decide If You Should Keep It
The recent controversy is less about size and more about expectations. The Gemini Nano model has hovered around 4GB since its 2024 debut, and the way Chrome decides to install it has not suddenly ballooned. What did draw fresh attention were wording changes in Chrome’s AI settings around version 148. Earlier, Chrome explicitly stated that on-device AI would not send data to Google’s servers; that line was removed, prompting questions about what might have changed under the hood. Google has said the underlying behavior remains the same, but the weaker language has made some users more cautious. If you value local AI features and understand their trade-offs, keeping the model may be worthwhile. If you prefer minimal software, maximum transparency, and tight storage budgets, disabling on-device AI and removing Gemini Nano is a reasonable, informed choice.