Errors

Standard Compute uses standard HTTP status codes. Error bodies follow the format of the API you called.

Error format

On OpenAI endpoints:

{
  "error": {
    "message": "Your monthly compute budget is used up. ...",
    "type": "insufficient_quota",
    "code": "budget_exhausted"
  }
}

On the Messages API:

{
  "type": "error",
  "error": { "type": "rate_limit_error", "message": "..." }
}

Status	Code	Meaning	What to do
`400`	`model_not_found`	The model ID doesn't exist.	Use `standardcompute` or an ID from `/v1/models`.
`400`	`invalid_model`	`model` is null, empty or not a string.	Send a model ID, or omit the field.
`400`	`unsupported_model_input`	Images were sent to a text-only model.	Use a vision model or `standardcompute`.
`400`	`unsupported_parameter`	A parameter isn't supported, such as `previous_response_id`.	See the endpoint's page.
`401`		The API key is missing or invalid.	Check the `Authorization` header.
`402`	`budget_exhausted`	Your monthly budget is used up.	Wait for the date in `x-sc-budget-renews-at`, or add budget.
`402`	`free_quota_exhausted`	Your free trial credit is used up.	Choose a plan.
`402`	`payment_required`	A payment failed and access is paused.	Update your payment method in Billing.
`403`	`company_access_denied`	This company key has been disabled.	Ask your company admin.
`429`	`rate_limit_exceeded`	Too many requests.	Wait for `retry-after` seconds, then retry.
`503`	`service_unavailable`	The model is temporarily unavailable after retries.	Retry after `retry-after`, or use smart routing.

Other 4xx errors from the model provider are passed through with provider details removed.

Standard Compute already retries provider errors before responding. When you retry:

Retry 429 and 503 after the retry-after delay, with exponential backoff.
Don't retry 400, 401, 402 or 403 without changing the request or your account.

Header	Description
`retry-after`	Seconds to wait before retrying, on `429` and `503`.
`x-sc-budget-renews-at`	When your budget resets, on `budget_exhausted`.
`x-sc-gated`	Why a request was blocked, for example `budget_exhausted` or `rpm_limit`.