Chrome and Edge on-device AI deployment

Chrome ve Edge’deki entegre yapay zeka modellerinin indirilebilmesi için en az 20 GB boş alan talep edilmeye başlandı

No separate article for this desk or language yet.

Google introduced a feature in February allowing users to disable and delete the on-device AI model from Chrome settings.

Around February 2026, Google introduced a feature in Chrome settings that lets users disable and delete the on-device AI model directly from the browser. Google spokesperson Scott Westover said that when the model is disabled, download and update processes are terminated.

The feature works through an "On-device AI" switch inside Chrome settings. When disabled, the browser deletes already downloaded models and stops new downloads; if settings are changed again, Chrome rescans hardware capacity and re-downloads the necessary files to reactivate the system.

The move came after reports that Chrome had been automatically downloading a file named 'weights.bin' — roughly 4 GB in size and described as fundamental to the Gemini Nano AI model — without any consent, notification, or option to disable it. Reports also said Chrome would secretly re-download and reinstall the file even if users manually deleted it.

Google has said Chrome has been using a Gemini Nano AI model running locally on systems since 2024, pointing to security features such as fraud detection and developer APIs that do not send data to the cloud.

Google began using a locally running Gemini Nano AI model in Chrome since 2024.

Google confirmed that Chrome has been using the Gemini Nano AI model, which runs locally on systems, since 2024. According to a Google spokesperson, this local model provides infrastructure for essential security features, such as fraud detection and developer APIs, without the need to send data to the cloud.

The implementation involves the use of a fundamental file named 'weights.bin' for the model. While the model file itself is approximately 4 GB, Google requires at least 20 GB of free disk space for on-device AI features to be used efficiently, accounting for post-installation updates, data processing, and model stability.