Users with a Google AI Pro plan are seeing Antigravity IDE lock them out with a “Model quota limit exceeded” message for several days even though the documented maximum is 5 hours. New chats show a 4 day lockout and existing sessions show a 7 day lockout.

Antigravity IDE shows multi day quota lockout on Google AI Pro accounts

You sign in with a valid Pro subscription, open or start chats that use Claude Opus 4.5 Thinking, and the IDE responds with Model quota limit exceeded and a wrong resume date such as 1 slash 16 slash 2026 for existing sessions and 1 slash 13 slash 2026 for new chats. This happens on Windows 11 in Google Chrome and only inside Antigravity IDE. For more context on what this error means and how it is computed, check our short AI guide on Antigravity quota behaviors: critical quota error reference.

Solution Overview

Aspect	Detail
Root Cause	Stale or corrupted usage cache and entitlement state inside the IDE, combined with a per model throttle on Claude Opus 4.5 Thinking that inflates the lockout window for specific conversations
Primary Fix	Fully refresh account entitlements and clear site storage for the IDE, then start a fresh chat on a different model and switch back if needed
Complexity	Easy
Estimated Time	10 to 15 minutes

Fix the incorrect 4 to 7 day lockout now

Step by Step Solution

Step 1. Sign out of Antigravity IDE and close all tabs

• Sign out from your profile inside the IDE
• Close every browser tab that points to the IDE

Step 2. Clear site storage for the IDE origin

• In Chrome type chrome colon slash slash settings slash siteData in the address bar
• Search for your Antigravity IDE domain and remove all data including Cookies, Local Storage, IndexedDB, and Service Workers
• Quit and relaunch Chrome

If you prefer DevTools 1. Open the IDE 2. Press F12 to open DevTools 3. Application panel then Storage 4. Click Clear site data Step 3. Re authenticate and confirm entitlements

• Sign back into the IDE with the Google AI Pro account
• Confirm the Pro badge or plan indicator shows correctly in the workspace header

To double check the subscription is active in your Google account you can also review it at My Subscriptions. Step 4. Start a brand new chat on a different model

• Create a new conversation instead of reopening the old locked one
• Choose a different model first, then switch back if required

– Example switch from Claude Opus 4.5 Thinking to a non thinking variant or another allowed model, then back to your preferred model

• Keep a single browser tab open and avoid parallel runs for the first message

Tip If your model picker exposes thinking configuration reduce thinking tokens for the first request For additional tactics to keep your usage below throttles, see our short checklist on managing Antigravity model triggers: understanding Antigravity quota triggers. Step 5. Verify system clock and time zone Wrong local time can confuse rolling windows shown in client messages even if the back end is correct. On Windows 11 1. Settings then Time and Language then Date and time then enable Set time automatically and Set time zone automatically 2. Sync time PowerShell commands if you prefer terminal

# View the current time zone
Get-TimeZone

Optionally set a specific time zone name
Set-TimeZone -Name "Pacific Standard Time"

Force time sync
w32tm /resync

Step 6. Cancel queued or stuck runs if your IDE shows a queue If the IDE has a Runs or Queue panel cancel all pending items so the next request starts clean. Step 7. Test with a lightweight prompt Send a short message first such as “Reply OK only” If this succeeds your quota window is refreshed. Then proceed with normal prompts. Step 8. If lockout persists capture evidence and request a quota reset Collect these in Chrome 1. Press F12 and open the Network tab 2. Reproduce the error once 3. Click the failed request then copy Request URL, Response, and the x request id or similar header 4. Save all as HAR with content Attach to your support ticket along with

• Screenshot of the lockout banner that shows the incorrect resume date
• Your account email and workspace id
• Approximate time of the error
• The model name and it was thinking enabled

While you wait you can keep working through a direct API where your paid key applies. Example calls below. Anthropic API example for a light test

curl https://api.anthropic.com/v1/messages \
 -H "x-api-key: $ANTHROPIC_API_KEY" \
 -H "anthropic-version: 2023-06-01" \
 -H "content-type: application/json" \
 -d '{
 "model": "claude-3-5-sonnet-latest",
 "max_tokens": 64,
 "messages": [{"role":"user","content":"Reply OK only"}]
 }'

Google Generative Language API example

curl -s -X POST \
 -H "Content-Type: application/json" \
 "https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-pro:generateContent?key=$GOOGLE_API_KEY" \
 -d '{ "contents": [{"parts":[{"text":"Reply OK only"}]}] }'

API limits for Google models are documented at Google AI pricing and quotas.

Alternative Fixes and Workarounds

Switch off thinking for the same model

• If your plan allows the base model without thinking use that for the session
• Thinking requests often count heavier against throttles

Create a fresh workspace or project

• Some quota flags are cached per project
• A brand new workspace can clear the inflated window immediately

Reduce concurrency

• Keep a single chat active
• Avoid sending multiple large messages in quick succession
• Wait one minute between heavy generations to allow counters to settle

Use a different browser profile

• Create a clean Chrome profile then sign in once and retry
• This isolates corrupted local storage that can cling to old quota windows

Temporary provider switch

• If your IDE supports multiple providers switch to a non affected model family for urgent tasks

If you need a deeper walkthrough of techniques that keep Antigravity below soft and hard limits, see our compact playbook here: strategies to avoid Antigravity throttles.

Troubleshooting Tips

• Confirm that the wrong resume date appears only in the IDE. If direct API calls succeed the issue is client side caching or entitlement propagation inside the IDE
• Check for mixed sign in. Verify that the Google account in Chrome is the same one used for Pro billing
• Disable extensions that intercept requests such as script blockers, privacy filters, corporate proxies
• Try a different network to rule out enterprise egress that may collapse many users into one IP level quota
• If you see different resume dates between new and existing chats it is strong evidence of per thread quota metadata. Always start a new chat after the reset steps above

Best Practices

• Keep thinking requests small and infrequent when using models with higher quota weight
• Avoid opening many parallel chats for the same model in multiple tabs
• Refresh site storage only for affected origins rather than clearing the entire browser
• Periodically verify your subscription and workspace entitlements after plan changes
• Capture request ids whenever an error appears. It accelerates support diagnostics

Final Thought

The multi day lockout is almost always a cached entitlement and per thread throttle bug, not a real quota limit. A clean sign out, targeted site storage reset, and a new chat on a different model restores access in minutes so you can get work done again.