Why Did Google Chrome Download a 4GB AI Model to Your Mac?

What Is Gemini Nano Doing Inside Google Chrome?

Gemini Nano is Google's smaller language model designed to run directly on a device rather than sending every AI task to a remote server. Chrome uses on-device models to support features such as writing assistance and text rephrasing, scam detection and security-related features, summarising content, and other browser features that use built-in AI capabilities.

The exact features available depend on your Chrome version, device, and feature availability. Unlike a cloud-based AI service, an on-device model needs its model files stored locally. These files contain the numerical parameters, commonly called weights, that the model uses to process information. That's why Chrome needs several gigabytes of storage for Gemini Nano.

Running a model locally can reduce the need to send certain processing requests to a server and may improve response times. But there's a trade-off: the model occupies space on your Mac even when you're not actively using an AI feature.

Google says Chrome only downloads these models on eligible devices. Its documented requirements include sufficient hardware performance, an unmetered network connection, and approximately 20 GB of free disk space. Requirements can change.

Read more: Chrome for Developers: Understand built-in model management in Chrome · Snopes Fact Check: Google Chrome silently downloads 4GB AI model

What Is the 4GB weights.bin File?

The file attracting attention is named weights.bin. It is stored inside a Chrome data directory called OptGuideOnDeviceModel. This directory contains the on-device model used by Chrome's optimisation guide infrastructure.

On macOS, the commonly reported location is: ~/Library/Application Support/Google/Chrome/OptGuideOnDeviceModel/. You may find a version-numbered subfolder inside it containing weights.bin. The exact directory structure can vary between Chrome versions and profiles, so don't assume every Mac will have precisely the same folder layout. The file can occupy around 4 GB, although its actual size may differ as Google updates the model.

How to find it using Finder:

1. Open Finder.
2. Press Command + Shift + G.
3. Paste this path: ~/Library/Application Support/Google/Chrome/
4. Look for OptGuideOnDeviceModel.
5. Open the folder and check for a version-numbered directory containing weights.bin.
6. If you find it, select the file and press Command + I to open Get Info. This shows its actual size.

If the folder doesn't exist, Chrome may not have downloaded the model to your current user profile.

Read more: Reddit: PSA Chrome silently downloaded a 4GB AI model on Mac

How to Check the Model's Size Using Terminal

You can also inspect the directory without opening individual folders. Terminal allows you to measure and find the exact file without deleting anything.

Open Terminal and run:

This reports the total size of the model directory if it exists. To locate weights.bin anywhere inside Chrome's user data directory, use:

These commands only inspect files. They don't delete or modify anything. If you use multiple Chrome profiles, remember that Chrome's user data can contain more than one profile. The model directory may not necessarily be inside the Default profile folder.

Why Deleting weights.bin Alone May Not Work

It seems straightforward: find the large file, move it to Trash, and reclaim the storage. But that doesn't address why Chrome downloaded it.

Chrome manages its on-device AI models automatically. If the on-device AI setting remains enabled and a feature needs the model, Chrome may download the missing files again. Google's own documentation explicitly warns that manually locating and deleting model files is not the supported way to manage them. This explains why some users report that the folder or file returns after they restart Chrome.

The important distinction is between removing a model file and disabling the feature responsible for managing that model. If you want to keep Chrome's on-device AI features, leave the model installed. If you don't want Chrome to download or maintain these models, use its supported setting instead.

Read more: Google Chrome Help: Manage on-device generative AI models in Chrome

How to Stop Chrome From Downloading Gemini Nano

Google now documents a setting that lets users manage on-device generative AI models through Chrome Settings.

Follow these steps:

1. Open Google Chrome.
2. Click the three-dot menu in the top-right corner.
3. Open Settings.
4. Select AI innovations.
5. Find On-device AI.
6. Turn the setting off.

The exact menu availability and wording can vary by Chrome version and rollout. According to Google's support documentation, turning this setting off: deletes the on-device generative AI model files, prevents Chrome from downloading these models again while the setting remains off, and disables features that depend on those on-device models. Some Chrome AI features don't rely on these generative models and may continue working.

What if you can't find the setting? Check whether Chrome is up to date and look again under Settings → AI innovations. The setting may not be available in every version, configuration, or managed browser installation. If your Mac is managed by an organisation, Chrome Enterprise policies may control these features. In that situation, contact your administrator rather than trying to override managed settings.

Avoid relying on random Chrome flags or terminal commands found in forum posts as a permanent solution. Experimental flags can change or disappear, and they aren't Google's documented model-management method.

Read more: Google Chrome Help: Manage on-device generative AI models in Chrome

Does Turning Off On-device AI Affect Chrome?

Yes, but only features that depend on the on-device generative models.

For example, you may lose access to certain local writing assistance or AI-powered features that require Gemini Nano. There is also a potential benefit: Chrome no longer needs to maintain those model files while the setting is off. You don't need to uninstall Chrome, clear your browsing history, or reset your browser to manage this storage. If you regularly use Chrome's local AI features, keeping the model may be worth the storage cost. If you never use them, disabling on-device AI can be a reasonable choice.

Why Does Chrome Need So Much Free Storage to Download an AI Model?

Google's documentation lists approximately 20 GB of available disk space as a requirement for downloading on-device generative AI models.

That doesn't mean Gemini Nano itself occupies 20 GB. The model is much smaller, with the commonly reported weights.bin file occupying around 4 GB. The additional free-space requirement gives Chrome room to manage downloads and other operations without immediately exhausting available storage.

Chrome also manages model updates in the background. Google's developer documentation explains that model updates can involve downloading a complete new model rather than applying a small incremental patch. This means temporary storage use can differ during an update. Chrome can also automatically remove the model when available storage falls below an internal threshold.

Read more: Chrome for Developers: Understand built-in model management in Chrome

Find Chrome's AI Model Alongside Other Large Files on Your Mac

The Gemini Nano model is just one example of an application storing large resources outside its main application bundle.

Other applications can also occupy substantial storage through:

• Browser caches and downloaded resources
• Local AI models
• Developer tool caches
• Application support files
• Virtual machines and container images
• Media libraries and temporary project files

The challenge is identifying which large files belong to which applications and understanding whether they're safe to remove. Diskfern helps you inspect your Mac's storage through Treemap, Sunburst, and List views. You can explore large folders, investigate application data, and identify storage consumed by developer tools and local AI models. Start with a free scan, locate the largest items, and inspect their locations before deciding what to remove.