CNET: Chrome may have quietly installed a ~4GB 'Gemini Nano' AI model on some devices

CNET reported on May 14 that some Google Chrome users may have found a large on-device AI model installed on their computers as Chrome expands Gemini-powered features. The report said the file can be about 4 gigabytes and may appear in Chrome’s on-device model files even if the user did not knowingly turn on an AI feature. Google’s public documentation confirms that Chrome can download Gemini Nano models for supported desktop systems and manage those downloads automatically in the background. Google has also said it will present Gemini updates at its I/O conference on May 19 and May 20 in Mountain View, California. ### Which Chrome feature is tied to the download? Google’s Chrome developer documentation says Gemini Nano is the language model used for several built-in AI APIs in Chrome, including the Prompt, Summarizer, Writer, Rewriter and Proofreader tools. Those APIs are designed for desktop systems, not Android or iOS, and Google says model handling is meant to be automatic for users and developers. (support.google.com) Google’s help page for Chrome says the browser may download on-device generative AI models in the background so features that rely on them remain ready for use. That support page also says deleting the models only disables features that depend on them, indicating the files are treated as part of Chrome’s feature set rather than as ordinary cached data. (developer.chrome.com) ### When does Chrome actually start downloading Gemini Nano? Google’s model-management documentation says the initial download is triggered by the first call to a `.create` function for an AI API that depends on Gemini Nano. The same page says Chrome downloads the model on demand after checking the device’s hardware and selecting a larger or smaller variant. (support.google.com) The same Google document says some `availability` checks can also trigger a model download shortly after a fresh profile starts if Chrome’s Gemini Nano-powered scam detection feature is active. Google says the download can continue if a tab is closed and can resume after a browser restart within 30 days. ### Which devices meet Google’s requirements? (developer.chrome.com) Google’s Prompt API documentation says supported systems include Windows 10 and 11, macOS 13 or later, Linux, and Chromebook Plus devices on supported ChromeOS versions. The page says the APIs that use Gemini Nano are not yet supported on Chrome for Android, iOS, or non-Chromebook Plus ChromeOS devices. (developer.chrome.com) Google’s published requirements also say the Chrome profile volume must have at least 22 GB of free space. The company says devices need either more than 4 GB of VRAM for GPU use or at least 16 GB of RAM and four CPU cores for CPU-based inference. Google adds that the exact model size can vary with updates and says users can check the installed version and size at `chrome://on-device-internals`. (developer.chrome.com) ### Can users remove the files or stop future downloads? Google’s Chrome Help page says users can delete on-device AI models, with the effect that features relying on those models will stop working. The same page says Chrome uses those models to power browser and web features, but the search result snippet available publicly does not spell out a single universal off switch for every Gemini-related feature. (developer.chrome.com) CNET said it published instructions earlier on May 14 for users who want to remove the model from affected systems. Google’s developer documentation also points users to `chrome://on-device-internals` to inspect the installed model and to Chrome’s AI feature controls and flags documentation for feature-level testing and configuration. ### Why is this surfacing now? (support.google.com) Google said in a February 17 post that I/O 2026 will run on May 19 and May 20 and will include updates across products “from Gemini to Android and more.” The company has been steadily expanding Gemini inside Chrome, including developer-facing built-in AI APIs and consumer-facing browser features. (support.google.com) Google’s next public checkpoint is May 19 at Shoreline Amphitheatre in Mountain View, where the company has said Gemini will be part of its I/O announcements. Users who want to verify whether a model is present before then can check Chrome’s on-device model page at `chrome://on-device-internals`, according to Google’s documentation. (blog.google)