If you want to stop Chrome downloading AI model files that quietly eat up 4GB or more on your Mac, you are not alone. Many users have noticed their MacBook storage shrinking after recent Chrome updates, only to discover Google has been silently downloading Gemini Nano and other on-device AI models in the background. These files power Chrome’s built-in AI features like Smart Compose, tab organisation, and on-page summarisation, but they are huge, persistent, and not always wanted. This guide walks you through exactly how to block, disable, and remove these AI model files from Chrome on macOS in 2026.
Why Chrome Is Suddenly Hogging 4GB of Your Mac’s Storage
Starting in late 2024 and accelerating through 2026, Google began rolling out Chrome built-in AI to all stable channel users. The browser now ships with a local AI runtime that downloads several large model files — typically Gemini Nano weights — directly onto your Mac. These models live in your user library and can grow to 4GB or larger depending on which features get activated.
The reason Google does this is to enable offline, low-latency AI features without sending every prompt to the cloud. While that sounds great in theory, the download happens silently, often over metered or slow connections, and rarely asks for your permission. For users with 256GB MacBook Airs or older Intel Macs, this can be a serious problem.
The chrome ai model storage footprint also includes additional helper models for translation, summarisation, and writing assistance. Each feature can pull its own multi-gigabyte payload. Worse, even if you never use these features, the models often remain on disk indefinitely.
Where Chrome Stores AI Model Files on macOS
On a Mac, Chrome stashes its AI assets inside the application support folder. The main path to know is ~/Library/Application Support/Google/Chrome/OptimizationGuidePredictionModels and the related OnDeviceModel directory. Inside, you will see folders with version numbers and large .bin or .tflite files.
You can verify the size yourself by opening Finder, pressing Cmd + Shift + G, and pasting that path. Sort by size and you will quickly see which sub-folders are responsible for the bloat. This is the same data we will be targeting in the cleanup steps below.
How to Disable Gemini Nano in Chrome Before It Downloads
The cleanest fix is to disable Gemini Nano Chrome features before the browser even attempts the download. Chrome exposes several experimental flags and enterprise policies that control on-device AI behaviour. Turning these off prevents new model fetches and stops background updates.
Follow these steps to disable the AI components from inside Chrome itself:
- Open Chrome and type chrome://flags into the address bar, then press Return.
- In the search box, type optimization guide on device.
- Set Enables optimization guide on device to Disabled.
- Search for prompt API for Gemini Nano and set it to Disabled.
- Search for summarization API, writer API, and rewriter API — disable each one.
- Click the Relaunch button at the bottom to restart Chrome with the new settings.
Pro tip: If you don’t see these flags, your Chrome version may have moved them to a different name. Try searching for on-device model or built-in AI instead. Google rotates flag names every few releases.
Blocking Downloads via Chrome Enterprise Policy
For a more permanent fix that survives Chrome updates, you can apply an enterprise policy on your Mac. This works even on personal machines and gives you bullet-proof control over the AI features.
- Open the Terminal app from Applications > Utilities.
- Paste this command and press Return: defaults write com.google.Chrome GenAILocalFoundationalModelSettings -integer 1
- Run a second command: defaults write com.google.Chrome OptimizationGuideOnDeviceModelEnabled -bool false
- Quit Chrome completely (Cmd + Q) and reopen it.
- Visit chrome://policy to confirm both policies are listed as active.
The integer value of 1 tells Chrome never to download the foundational model. This is the same flag IT administrators use in managed Mac fleets, and it works perfectly on consumer devices too.
How to Remove Chrome AI Files That Are Already on Your Mac
Disabling the flags stops new downloads, but it does not always delete the gigabytes already sitting on disk. To remove Chrome AI files properly, you need to manually clear the model directories. This is safe — Chrome will simply skip re-downloading them once the policies above are in place.
Here is the recommended cleanup procedure:
- Completely quit Chrome (Cmd + Q, not just close the window).
- Open Finder and press Cmd + Shift + G.
- Paste: ~/Library/Application Support/Google/Chrome/
- Locate and delete the OptimizationGuidePredictionModels folder.
- Delete the OnDeviceModel folder if present.
- Empty the Trash to actually reclaim the space.
- Reopen Chrome and verify in chrome://components that Optimization Guide On Device Model shows version 0.0.0.0.
Warning: Do not delete the entire Chrome folder unless you want to lose bookmarks, passwords, and extensions. Only target the AI-specific sub-folders listed above.
Using a Storage Audit to Catch Hidden AI Caches
Sometimes Chrome leaves behind orphaned model fragments in less obvious locations. A quick chrome storage cleanup mac audit using the built-in Storage Manager can surface these. Click the Apple menu, choose About This Mac, then More Info and Storage Settings.
Look for Chrome under the Applications or Documents category. If it shows anything dramatically larger than the app itself (around 600MB), there are still model files hiding somewhere. You can also use a free disk visualiser to walk the file tree and spot multi-gigabyte binaries. For broader Mac tuning advice you can browse the Hawkdive technology tutorials hub for related guides.
Stopping Chrome from Re-Downloading the Models Later
One frustrating discovery many users make is that Chrome will silently re-download the models after a major update, even if you previously deleted them. To prevent this from happening again, you need a belt-and-braces approach combining flags, policies, and a network-level safeguard.
- Keep the enterprise policy active — do not reset Chrome to defaults after updates.
- Disable background updates for the optimisation guide component in chrome://components.
- Block the model CDN at your router or in your hosts file if you want absolute certainty (advanced users only).
- Check chrome://components monthly to confirm the on-device model still reads 0.0.0.0.
- Avoid clicking any in-browser prompts that ask to enable AI writing tools or smart suggestions.
If you accidentally re-enable any AI feature through a tempting pop-up, the entire model download will start again within minutes. Be especially vigilant after Chrome’s quarterly feature drops, which often introduce new AI surfaces in places like the address bar and tab manager.
Alternative Browsers if You Want a Lighter Experience
If Chrome’s AI bloat keeps coming back despite your best efforts, you might consider browsers that take a more restrained approach. Safari on macOS handles AI features through Apple Intelligence at the OS level, meaning the browser itself stays lean. Firefox also keeps its AI features opt-in and modest in size.
That said, switching browsers is a big move. Most readers find that disciplined flag management is enough. If you are exploring AI tools more broadly, our Perplexity AI app review for 2026 compares how different assistants handle local versus cloud processing.
Other Storage and Performance Gotchas Mac Users Should Know
Chrome’s AI models are not the only sneaky storage hog on modern Macs. Many apps now ship with bundled machine-learning components that quietly inflate over time. WhatsApp media caches, Photos analysis databases, and even Mail’s on-device categorisation can each consume several gigabytes.
If you’ve already cleaned up Chrome and still feel cramped on storage, take a look at our guide on the WhatsApp local storage bug on iPhone and how to free up space fast. The same principles apply when audit-cleaning bloated app data on Mac.
Security-conscious users should also be aware that local AI models can occasionally trip vulnerability scanners. We’ve covered known cases in our breakdown of Anthropic AI vulnerability scanner issues on macOS, which contains useful context on how on-device AI interacts with system security tools.
Key Takeaways Before You Move On
- Chrome silently downloads up to 4GB of AI model files on macOS by default.
- Disabling the relevant flags in chrome://flags stops new downloads immediately.
- Enterprise policies via Terminal offer the most reliable long-term block.
- Manually deleting the OptimizationGuidePredictionModels folder reclaims the space.
- Re-check after every major Chrome update to make sure nothing has re-enabled itself.
Frequently Asked Questions
Why is Chrome downloading 4GB of AI files?
Chrome downloads these large files to power its built-in AI features like Gemini Nano, on-device summarisation, writing assistance, and translation. Google pushes the models to enable offline, low-latency AI without round-trips to the cloud. The download happens automatically once your Chrome version supports the feature, regardless of whether you actually use it.
How do I disable Chrome’s built-in AI features?
Open chrome://flags and disable any flag containing optimization guide on device, Gemini Nano, or the various API names like writer, rewriter, and summarisation. For a more permanent solution, apply the OptimizationGuideOnDeviceModelEnabled policy via Terminal as described above. Relaunch Chrome and the AI surfaces will disappear from menus and the address bar.
Can I delete Chrome’s AI model files safely?
Yes, deleting the OptimizationGuidePredictionModels and OnDeviceModel folders inside Chrome’s Application Support directory is completely safe. Chrome will simply behave as if the models were never downloaded. Your bookmarks, passwords, history, and extensions remain untouched because they are stored in different folders entirely.
Does Chrome AI work offline on Mac?
Yes — that is precisely why the models are so large. Once downloaded, Gemini Nano and the associated APIs can summarise pages, draft text, and answer prompts entirely on-device with no internet connection required. The trade-off is the substantial disk footprint, which is the main reason many users prefer to disable the feature entirely.
How much storage does Chrome AI use?
A typical install consumes between 2GB and 4GB once all on-device models are downloaded, but this can climb higher as Google ships additional capabilities. The Gemini Nano weights alone are roughly 2GB, with extra payloads for language translation, summarisation, and image processing adding another 1–2GB on top.
Final Thoughts
Learning how to stop Chrome downloading AI model files on your Mac is one of the highest-impact storage wins available in 2026. With a few flag changes and a Terminal command, you can reclaim multiple gigabytes and prevent the browser from quietly re-downloading them after every update. Keep an eye on chrome://components each month and you’ll stay in control of your disk space for good.
If this guide helped, you may also find value in our walkthrough on fixing the Gmail unreadable emails bug on Android 16 or our practical write-up about Instagram accounts hacked via the Meta AI chatbot and how to recover them. Both cover real-world AI-era issues affecting users right now.
