Limits & Errors

Request size limits, unsupported features, and error responses of the Inference API.

Private beta: Email support@sliplane.io to request Inference access for your team.

Request limits

Endpoint	Maximum request body
`/v1/chat/completions`	32 MiB
`/v1/responses`	32 MiB
`/v1/embeddings`	4 MiB, up to 128 inputs
`/v1/audio/transcriptions`	25 MiB, including the file and form fields
`/v1/systemone`	12,000 bytes, 1–32 questions

Responses: stored state (store must be false), background requests, saved prompts, and uploaded file references. Send the conversation history with each request. Tools must be function or custom tools executed by your client.
Embeddings: streaming and token ID inputs.
Transcription: streaming and confidential keys.
Decisions: images, streaming, and extra request fields.

Errors use the OpenAI error format:

{
  "error": {
    "message": "API key spend limit reached.",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }
}

type is invalid_request_error for 4xx responses and server_error for 5xx responses.

Status	Meaning
`400`	The request body or parameters are invalid
`401`	The key is missing, invalid, deleted, or lacks the `inference` scope
`404`	The model is unknown or not available on this endpoint or for this key
`413`	The request body exceeds the limit
`429`	The key reached its spend limit
`502`	The model returned an invalid response
`503`	The service is temporarily unavailable
`504`	The model did not respond in time

Retry 502, 503, and 504 with backoff.

Once a stream has started, the HTTP status stays 200. Errors after that point arrive as an event in the stream.