Chrome AI troubleshooting

What the status messages mean

The extension asks Chrome one question — is the on-device model usable right now? — and shows you the answer verbatim. There are four:

Translator files are separate downloads. Seeing translators as downloaded and available while the language model says downloading is normal and tells you nothing about Gemini Nano.

"Downloading" that never finishes (or sits at 0.0%)

This is the most common Chrome AI problem, and it is a Chrome-side stall: Chrome reports downloading, but no data is being transferred and the download never completes. Chrome only reports download progress to whichever page or extension started the download, so a download owned by something else looks like 0% forever from here. The extension now falls back to reporting Chrome's own status instead of showing a frozen percentage, but it cannot make the download move.

Work through these in order — stop as soon as Chrome AI reports ready:

1. Quit Chrome completely and reopen it. Not just the window — close every Chrome window, and check the system tray / Task Manager (Windows) or Force Quit (macOS) for a lingering Chrome process. A wedged on-device model service lasts as long as the browser process does, and this alone fixes many cases.
2. Force a component refresh. Open chrome://components, find Optimization Guide On Device Model, and click Check for update. Wait for the version line to change or say Up to date, then reopen the extension.
3. Look at what Chrome thinks it is doing. Open chrome://on-device-internals → Model status. It shows the installed version, criteria, and whether a fetch is genuinely running. The Event logs tab shows failures Chrome does not surface anywhere else.
4. Reinstall the model. Follow reinstall the on-device model below. This is the reliable fix when Chrome is holding a half-installed or superseded model version.

After each step, re-run the built-in test further down this page — it asks Chrome directly, so you get an answer without hunting through the extension. Between attempts, close and reopen the extension popup too, so it re-checks rather than showing a cached answer.

It worked before, now it wants to download again

Chrome ships new versions of the on-device model independently of the extension, and it can keep more than one version on disk at a time. When Chrome switches to a newer base model, the model your analyses used yesterday is superseded, and the API reports downloading until the new one is fully installed and validated. Two things follow from that:

• A re-download after a Chrome update is expected behavior, not a sign anything is broken. The weights are multiple gigabytes, so it is not instant.
• If Chrome ends up pointing at a version whose supporting files never finished installing, it can sit in downloading indefinitely. That is the stall described above, and a reinstall clears it.

Nothing is lost either way — Chrome AI is one model choice among many, and your lenses, history, and profile are unaffected.

"Model not downloaded yet" — how to start it

Chrome requires a real click before it will download the model — an extension cannot start it silently. In SoWhatify:

1. Open the extension, then Edit profile & API → API tab.
2. Select a Chrome AI model, then click Start Chrome AI.
3. Leave the tab open. The first download is several gigabytes and continues in Chrome even if you close the popup.

"Not available" or "not supported on this device"

• Chrome version and platform. Chrome AI needs Chrome 138 or newer on desktop — Windows, macOS, or Linux. Chrome on Android and iOS, and most Chrome-derived browsers, do not expose it. Check chrome://settings/help.
• AI features switched off. Open Chrome Settings → System (or search for AI in settings) and make sure on-device AI features are enabled, then reopen the extension.
• Flags. On some builds the Prompt API still needs Prompt API for Gemini Nano enabled in chrome://flags, followed by a restart.
• Managed devices. A work or school policy can disable on-device AI outright. If chrome://on-device-internals reports the feature as disabled and you cannot change it, use a cloud model.
• Hardware. Gemini Nano needs a reasonably capable GPU and enough VRAM. Older or low-end machines report unavailable permanently, and no amount of reinstalling changes that.

Requirements and limits

• Chrome 138+ on desktop (Windows, macOS, Linux).
• Roughly 22 GB free disk space on the drive holding your Chrome profile. Chrome refuses to start the download below that, even though the model itself is smaller.
• An unmetered network connection, and not running in battery-saver mode.
• A supported GPU with enough VRAM.
• Chrome AI has a much smaller context window than cloud models, so very long articles are trimmed more aggressively than they would be with a cloud model.

Test Chrome AI right here

This asks your browser the same questions the extension asks, then runs a real prompt through the on-device model. Everything happens locally in this tab — the prompt and the model's answer never leave your computer, and this site's server never sees them.

Compare the results with the status table. If this page also reports downloading, the extension is relaying Chrome faithfully and the fix belongs in Chrome — start with the stall checklist. If the language model API is missing entirely here, see not available.

One caveat: this page tests the API as a web page sees it. Chrome exposes the same on-device model to extensions on a slightly different schedule, so if this page says the API is unavailable while the extension works fine, believe the extension.

console.log("LanguageModel:", await LanguageModel?.availability());
console.log("Summarizer:", await Summarizer?.availability());

Reinstall the on-device model

Try the built-in uninstall first; it is the least invasive.

Method 1 — uninstall in Chrome internals

1. Open chrome://on-device-internals.
2. Open the Model status tab.
3. Click Uninstall.
4. Restart Chrome.
5. In the extension: Edit profile & API → API → Start Chrome AI.

Method 2 — remove the model folder by hand

Use this if Method 1 is unavailable or does not stick. The folder is Chrome-managed cache data that Chrome re-downloads; your bookmarks, profile, and extension data live elsewhere and are not touched.

1. Quit Chrome completely, including any tray or background process.
2. Delete the OptGuideOnDeviceModel folder in your Chrome data directory:
   • Windows: %LOCALAPPDATA%\Google\Chrome\User Data\OptGuideOnDeviceModel
   • macOS: ~/Library/Application Support/Google/Chrome/OptGuideOnDeviceModel
   • Linux: ~/.config/google-chrome/OptGuideOnDeviceModel
3. Start Chrome again.
4. In the extension: Edit profile & API → API → Start Chrome AI, and leave the tab open while it downloads.

Afterwards, confirm in chrome://on-device-internals that the model reports as ready and the crash count is not climbing.

Use a cloud model instead

Chrome AI is optional. Every lens and chain works with a cloud model, usually faster and with a much larger context window. In the extension, open Edit profile & API → API, paste a key for OpenAI, Anthropic, Gemini, or OpenCode, and pick one of its models in the popup. You can switch back to Chrome AI whenever the download sorts itself out.

Still stuck? Send us a ticket

Open a support ticket and include:

• Your Chrome version, from chrome://settings/help.
• Your operating system.
• The exact status text the extension shows.
• The two console lines from the check above.
• What the Model status tab in chrome://on-device-internals says.

Never include API keys in a ticket. We only need the status output, not your credentials.