I tried to run a prompt with Claude 5.5, the system threw an Error ID, did absolutely no task, but deducted my credits from 3000 down to 1300. Please check logs and refund my balance
Normally, a single model request that errors out immediately isn’t charged. However, when Use AI Credits is turned on, credits are metered per token on each step the agent takes. With Claude 5.5 (especially Claude Opus 5.5 with thinking enabled, or in a long chat thread with a large context window), the agent can complete an initial thinking or file-reading step—which deducts credits for those tokens—and then hit a stream timeout or backend error on the next step before finishing your task.
Because we can’t access your Google One billing ledger or issue credit refunds directly from a public forum post, here is the fastest way to get your balance refunded and have us check the logs:
- Share the Error ID here so we can check the server logs: Please reply with the exact Error ID that popped up, the approximate time it happened (with your timezone), and whether you were using Claude Opus 5.5 or Sonnet 5.5.
- Request a credit refund through Google One Support: Since AI Credits are billed and managed through Google One, please reach out to Google One Support (or use the support/billing link inside the
/creditsmenu or Settings → Models & Usage) and include that Error ID so the billing team can credit your balance back. - Prevent unexpected credit drain in the meantime:
- You can pause automatic credit fallback anytime in Settings → Models & Usage (toggle off Use AI Credits), or in the CLI by running
/settingsand setting Use G1 Credits to off ("useG1Credits": false—see the AI credits guide). - Starting a new conversation (
+) before running heavy Claude 5.5 prompts also helps prevent large chat histories from multiplying token costs on each step.
- You can pause automatic credit fallback anytime in Settings → Models & Usage (toggle off Use AI Credits), or in the CLI by running