Skip to main content
Last Updated: September 23, 2026 Read the HTTP status and error code together when an AI Gateway request fails.

Pay-Per-Token Access Errors

For credit management, see Billing. For creating, replacing, or revoking a key, see Manage Your AI Gateway API Key. For the distinction between self-service pay-per-token access and closed-beta access, see Access Model.

Authentication Checks

For missing or invalid credentials, check all of the following without logging the key:
  • The base URL is https://ai.salad.cloud/v1.
  • The request uses Authorization: Bearer <your-ai-gateway-api-key>.
  • The value is the organization’s AI Gateway key, not the account’s Current API Key.
  • The key has not been deleted or replaced by a regenerated key whose overlap period has ended.
A successful GET /v1/models request confirms model discovery, not that your key or credit balance permits a billable completion.

Admission and In-Flight Requests

Key validity and pay-per-token access are checked when a request is admitted. A request rejected at admission for one of the access errors above is not processed as a completion. Once admitted, a request can finish and its tokens are billed even if its key is revoked or regenerated, or the organization’s credits run out before it finishes. This also applies to streaming requests. Deleting a key is not a way to cancel an existing completion or avoid its charge. See Requests Already in Progress.

Retry Safely

  • For credits_exhausted or no_access, resolve the billing or access issue before submitting again.
  • After a timeout, connection failure, server error, or partial stream, the outcome and charge can be uncertain. Do not automatically repeat the request; a new completion can incur another charge.
If the issue persists, contact SaladCloud Support with the endpoint, model ID, UTC time, HTTP status, error code, and request identifier if provided. Do not include API keys or sensitive prompts and outputs. Agents should follow the troubleshooting runbook.