Issue Description:
As a paying Pro subscriber, I rely on premium models for my workflow. However, requests sent to Claude 4.6 Opus, Claude 4.6 Sonnet, or Gemini Pro via the Antigravity agent consistently hang for about 5-6 seconds and then crash with an HTTP 500 Internal Server Error (ESF server, text/event-stream).
Additionally, on rare occasions, the UI shows the model “thinking” for about 3 seconds, my Pro usage limits/quotas are deducted, and then it immediately throws the exact same error. Losing paid quota for failed backend requests is extremely frustrating.
Ironically, the basic Flash model processes the exact same workspace without any issues.
Troubleshooting already done:
I have already tried creating an .agignore file for heavy directories, clearing IDE caches, and killing background processes. The premium models remain completely broken for workspace context tasks.
Trace IDs for Backend Engineers:
Here are the logs from the gateway timeouts:
Trajectory ID: 7e951708-27d8-4e26-8fd6-e637070c36.. | TraceID: 0x82a4463a0121e5..| Server-Timing: dur=5241Trajectory ID: de3fcb44-61cb-4cd2-b752-32da8fb8a5.. | TraceID: 0xff07f8f04e95f7..| Server-Timing: dur=5728
Please investigate these TraceIDs and fix the payload/timeout limits on the backend. Paying for a Pro subscription when core premium features constantly crash—and drain quotas in the process—is an unacceptable experience. Looking forward to a fix.
