Errors

Runars returns errors in the dialect of the endpoint you called. This page lists the shapes, including the ones that aren't obvious from the SDK defaults.

Error response shapes

Anthropic dialect (/v1/messages, /v1/messages/count_tokens)

{
  "type": "error",
  "error": {
    "type": "invalid_request_error",
    "message": "…"
  }
}

OpenAI dialect (/v1/chat/completions, /v1/responses, /v1/models)

{
  "error": {
    "message": "…",
    "type": "invalid_request_error",
    "code": "…",
    "param": null
  }
}

Error reference

401 — Unauthorized

Invalid, missing, or revoked API key. API-key revocation is cache-invalidated within a few seconds on Runars; if you just revoked a key, wait ~5s and the old key stops working.

402 — payment_required

Returned when your organization's balance is zero (or negative). Top up at Dashboard → Billing, or enable auto-reload to avoid this in production.

404 — model_not_found

Returned when the requested model id isn't in the catalog. Runars never silently substitutes another model—send a valid id from GET /v1/models. Example (OpenAI shape):

{
  "error": {
    "message": "model_not_found",
    "type": "invalid_request_error",
    "code": "model_not_found",
    "param": "model"
  }
}

429 — rate limited

Returned when you exceed your organization's rate-limit tier. Includes a retry-after header (seconds) indicating how long to wait before retrying.

5xx — upstream errors

A 5xx indicates an upstream or infrastructure failure. Idempotent requests can be retried with exponential backoff.

Retry guidance

• 401 / 404: do not retry—fix the key or model id first.
• 402: do not retry—top up the balance.
• 429: retry after the retry-after value, or respect the header.
• 5xx: retry with exponential backoff and jitter; cap at a few attempts.