I’ve been hitting “Agent terminated due to error” almost every time I pick a Claude model in Antigravity. Gemini works fine in the same session, so I dug into the logs instead of guessing.
Every single failure is the same line in ls-main.log:
UNAVAILABLE (code 503): No capacity available for model claude-opus-4-6-thinking on the server
I went back through my log sessions and counted:
- June 4: 56 errors in one session, 116 in another (that day it was hitting claude-sonnet-4-6 too, not just opus)
- June 6: 14
- June 7: 70
- June 11: 12
268 rejections across 5 sessions in 8 days. The agent typically runs for ~13 seconds, then dies. Retry does nothing or works once in twenty tries.
Before anyone suggests the usual: it’s not my quota (meter sits at ~100% because the requests never get served), not auth (login, loadCodeAssist and fetchAvailableModels all succeed right before the 503), not my setup (IDE v1.107.0 stable, macOS, and Gemini runs perfectly in the same window). This is the server saying no.
There are already threads about this going back months (122086, 167768, 119277) and the answers have been “peak hours, try later.” That’s not an answer when Claude access is a headline feature of the Ultra tier and it’s dead during normal working hours. I’m paying $200/month and routing my Claude work through other tools because the thing I paid for can’t serve a request.
What I’m asking for:
- Someone from the team to actually confirm Claude capacity for Ultra is under-provisioned, and when it gets fixed
- Reserved capacity or a queue for paying Ultra users instead of a hard 503
- Prorated credit or an extension for the affected period
I have full logs and trace IDs (e.g. 0x47f0f57d50d462..) if an engineer wants them.