How to fix Antigravity model list disappeared and stuck on loading models?

Antigravity shows “Pick a model” but the list is empty, or it hangs forever at “Loading models…”. You may also see agent terminations or capacity errors right before the list disappears.

How to fix Antigravity model list disappeared and stuck on loading models?

Users report the model selector is blank on Apple Silicon (M1/M3) and other Macs, and the UI stays at “Loading models…”. Some also hit capacity/overload or quota errors, then the selector vanished after a reconnect or app restart. On another machine the list might appear, but prompts won’t send—pointing to a mix of backend capacity issues, cached session problems, and network/DNS path issues.

Tip: For a deeper checklist on model availability issues, see our models not loading guide.

Solution Overview

If your agent keeps hanging in “loading” even after models reappear, also see these proven steps in our agent loading fixes.

How to fix Antigravity model list disappeared and stuck on loading models?

Step-by-Step Solution

1) Check live service status first

• If there’s an ongoing outage/capacity event, local fixes won’t help until it clears.
• Check:
• Google Cloud Status: https://status.cloud.google.com/
• Recent reports: PiunikaWeb roundup
• If an incident is active, the best action is to wait and retry later. If not, continue below.

2) Force a fresh models catalog fetch (sign-out, reload, sign-in)

• Sign out of Antigravity.
• Fully quit the app/browser.
• Reopen Antigravity and sign in again.
• Trigger a hard reload:
• macOS browser: hold Shift and click Reload, or use Cmd+Shift+R in Chrome/Edge.
• Confirm the selector populates.

3) Switch network path and flush DNS

Some users fix it instantly by changing the network (ISP → mobile hotspot). ISPs can cache or rate-limit routes.

• Try a different network:

• Disconnect Wi‑Fi → tether to your phone’s 4G/5G hotspot, or enable a reputable VPN.

• If models appear on the alternate network, the issue is routing/DNS on your primary ISP.

• Flush macOS DNS cache (safe, reversible):

• Open Terminal and run:

sudo dscacheutil -flushcache; sudo killall -HUP mDNSResponder

• Reconnect to your primary network and test Antigravity again.

• Optional: Temporarily change macOS DNS to public resolvers (revert later):

# Set DNS for Wi‑Fi to Google + Cloudflare
networksetup -setdnsservers Wi-Fi 8.8.8.8 1.1.1.1

# To revert to automatic (DHCP-provided) DNS:
networksetup -setdnsservers Wi-Fi "Empty"

4) Clear site/app cache and stale session data

Stale local storage can keep a bad “empty models” response.

• If using a browser:

• Chrome/Edge: Settings → Privacy → Cookies and other site data → See all site data → search for the Antigravity domain → Remove.

• Safari: Preferences → Privacy → Manage Website Data → search and remove the Antigravity entry.

• Close and reopen the browser, then sign in again.

• If using a desktop app:

• Look for an in‑app option like “Reset application data” or “Sign out everywhere.”

• Quit and relaunch the app after resetting.

5) Reauthorize and confirm plan/quotas

Capacity/quota errors can block or hide certain models.

• Sign out and sign back in to refresh tokens.
• Confirm your subscription/plan is active and your tenant/region allows the models you expect.
• Try a lighter model first (e.g., non‑Ultra) to see if the list loads.

6) Validate with the Gemini Models API (optional but precise)

If you have an API key, check if the model catalog is reachable from your machine. A 200/404 proves the path is live; timeouts indicate a network path problem.

• Quick reachability check:

curl -I -4 https://generativelanguage.googleapis.com

• List models with API key (expect HTTP 200 and JSON; replace YOUR_API_KEY):

curl -s -o /dev/null -w "%{http_code}\n" "https://generativelanguage.googleapis.com/v1/models?key=YOUR_API_KEY"

• If this only works on a different network (e.g., mobile hotspot), the issue is with your ISP/DNS/VPN route.

Alternative Fixes & Workarounds

• Use a different browser or Incognito/Private window

• Incognito bypasses some cached storage and extensions.

• Disable extensions and VPN/proxy temporarily

• Privacy extensions, corporate proxies, or strict VPNs can block catalog calls.

• Force IPv4

• Some ISPs have flaky IPv6 routes. Test with:

curl -I -4 https://generativelanguage.googleapis.com

• If IPv4 works but IPv6 fails, disable IPv6 on your adapter temporarily or keep a VPN that prefers IPv4.

• Wait and retry during low-traffic hours

• Capacity throttles are time-based; late evening spikes were frequently reported.

Working with agents specifically? You may benefit from our practical tips here: stuck agent troubleshooting.

Troubleshooting Tips

• Quota reached recently? If you saw “hit the model quota,” wait a short period, then try a smaller model. Quotas can block list population indirectly.
• Time and date accuracy: Ensure your system clock is correct; invalid TLS handshakes can block catalog fetches.
• Corporate network policies: Test outside VPN/proxy or on a personal connection to rule out SSL inspection and blocked endpoints.
• Try another machine/account: If it works elsewhere, export/import only your content and avoid copying browser profiles that may carry corrupt storage.
• Document the failure: Grab a timestamp and your public IP (whatismyip) to help support correlate with backend logs.

If your model list reappears but projects won’t open, this related walkthrough helps: fixing stuck Stitch projects.

Best Practices

• Keep the app/browser updated and clear site data periodically if you switch regions or plans.
• Avoid running many concurrent sessions/tabs for the same workspace.
• Use reputable DNS (Google or Cloudflare) and keep a fallback network (mobile hotspot) ready for outages.
• Bookmark a status page and test endpoint (curl to generativelanguage.googleapis.com) to separate local vs service-side issues.

Final Thought

In most cases the fix is quick: sign out/in, hard‑reload, and switch networks to force a clean model catalog fetch. If there’s a broader outage, the checks above confirm it fast so you can avoid chasing local ghosts and get back to work as soon as service stabilizes.