API reference

Errors & rate limits

Merius uses standard HTTP status codes and returns errors in the OpenAI shape, so your existing error handling works without changes. Each model has dedicated capacity; when one is briefly saturated we return a clean 429 in under a second rather than queuing you into a timeout.

Status codes #

Every response carries one of these status codes:

Parameter	Type	Description
`200`	OK	The request succeeded.
`400`	bad request	Malformed JSON or an invalid parameter. The error body says which field.
`401`	unauthorized	Missing or invalid API key. Check the Authorization header.
`402`	payment required	Insufficient credit balance. Add credits in the dashboard.
`404`	not found	Unknown model slug or endpoint. Check the model id and path.
`429`	rate limited	Model briefly saturated or your rate limit reached. Back off and retry; returned in under a second.
`500`	server error	An error on our side. Safe to retry with exponential backoff.

The error object #

Errors return a JSON body with an error object: a human-readable message, a type, and where relevant the offending param and a code. This matches OpenAI’s error shape:

Error body

{
  "error": {
    "message": "Unknown model: gpt-foo",
    "type": "invalid_request_error",
    "param": "model",
    "code": "model_not_found"
  }
}

Rate limits #

Limits are applied per key. When you exceed your limit, or a model is momentarily saturated, you get a 429 quickly so you can retry rather than blocking. Use exponential backoff with jitter on 429 and 500.

Retry 429 and 5xx responses with exponential backoff; do not retry 400, 401, or 402.
Spread bursts over time rather than firing every request at once.
Need higher, guaranteed throughput? A dedicated endpoint reserves capacity for your workload — see Two ways to run on the home page.

Live status and 90-day uptime history are on the status page, linked in the footer.