Chrome's Gemini Nano AI Model Consumes 4GB of Disk Space
According to a report from The Verge, Google Chrome may be silently using up to 4GB of local disk space for its on-device AI features. The culprit is a file named weights.bin located in the OptGuideOnDeviceModel directory under Chrome's data folder. This file contains the training parameters for Google's Gemini Nano model, which powers features like scam detection, writing assistance, autofill, and suggestion tools directly on your machine without cloud calls.
How to Check and Reclaim Storage
To see if the file is present on your system, navigate to your Chrome user data folder. The exact path varies by OS, but typically looks like:
- Windows:
%LOCALAPPDATA%\Google\Chrome\User Data\OptGuideOnDeviceModel - macOS:
~/Library/Application Support/Google/Chrome/OptGuideOnDeviceModel - Linux:
~/.config/google-chrome/OptGuideOnDeviceModel
If you delete the weights.bin file manually but keep AI features enabled, Chrome will re-download it on the next update. The correct way to free up the space permanently is to head to Settings > System and toggle off the On-Device AI option. This removes the model and disables AI features that rely on it.
Google's Statement and Caveats
Google spokesperson Scott Westover clarified: “We’ve offered Gemini Nano for Chrome since 2024 as a lightweight, on-device model. It powers important security capabilities like scam detection and developer APIs without sending your data to the cloud. ... The model will automatically uninstall if the device is low on resources.”
As of February 2026, Chrome provides a user-facing toggle to disable and remove the model. However, the 4GB size is only documented in a lengthy help center article, not at the point of enabling the features. Google notes that “Gemini Nano’s exact size may vary as the browser updates the model.”
If you rely on Chrome's AI features but are tight on storage, there is currently no option to switch to a cloud-based model; the only workaround is to disable on-device AI entirely.
📖 Read the full source: HN AI Agents
👀 See Also
Talkie: A 13B LLM Trained Exclusively on Pre-1931 Text, Using Claude as a Judge in RL Training
Researchers released Talkie, a 13B LLM trained only on text published before 1931 (no internet, no WWII data). Claude Sonnet 4.6 was used as the judge in its online DPO reinforcement learning pipeline, and Claude Opus 4.4 generated synthetic multi-turn conversations for fine-tuning. The model can write Python code from a few in-context examples despite zero modern code in training.
Claude Opus 4.6 and Sonnet 4.6 now offer 1M context at standard pricing
Claude Opus 4.6 and Sonnet 4.6 now include a full 1M context window at standard pricing with no long-context premium, plus expanded media limits to 600 images or PDF pages per request.
Claude Code v2.1.146: /code-review Command, Pagination Fix, Windows PowerShell Fix
Claude Code v2.1.146 renames /simplify to /code-review with optional effort level, fixes MCP pagination and Windows PowerShell tool, improves auto-updater reliability and diff rendering performance.
Proving Model Identity with Tinfoil's Modelwrap Technology
Tinfoil's Modelwrap ensures that inference providers serve the exact model weights they claim to, using cryptographic commitments verified by secure enclaves.