[BUG] 503 MODEL_CAPACITY_EXHAUSTED when using gemini-3-pro-low

[shirokun20]

When using antigravity-claude-proxy, requests routed through CloudCode intermittently fail with a 503 error indicating model capacity exhaustion.

Error Log

[CloudCode] Stream error at https://daily-cloudcode-pa.googleapis.com: 503 - {
  "error": {
    "code": 503,
    "message": "No capacity available for model gemini-3-pro-low on the server",
    "status": "UNAVAILABLE",
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.ErrorInfo",
        "reason": "MODEL_CAPACITY_EXHAUSTED",
        "domain": "cloudcode-pa.googleapis.com",
        "metadata": {
          "model": "gemini-3-pro-low"
        }
      },
      {
        "@type": "type.googleapis.com/google.rpc.RetryInfo",
        "retryDelay": "39s"
      }
    ]
  }
}

Steps to Reproduce

1. Send a request through the proxy targeting gemini-3-pro-low
2. Wait for streaming response
3. Occasionally receives 503 instead of output

Expected Behavior
Requests should either:

• Automatically retry after the suggested delay, or
• Fallback to another available model

Actual Behavior
Request fails immediately and must be manually retried.

Suggested Improvement
Implement automatic retry logic using RetryInfo.retryDelay, or allow configurable fallback models when capacity is exhausted.

Environment

• Proxy: antigravity-claude-proxy
• Endpoint: daily-cloudcode-pa.googleapis.com
• Model: gemini-3-pro-low

Additional Context
Error indicates server-side capacity exhaustion, but client-side handling could be improved for resilience.

[fenny-org]

Quick note: for 429s with MODEL_CAPACITY_EXHAUSTED you don't need to wait for retryDelay at all, you can keep spamming the endpoint until it goes through.