Developers

Keys and guides for the OpenAI- and Anthropic-compatible API. Requests are paid from the same balance as chat.

Error format

/api/v1/messages and /api/v1/messages/count_tokens return errors in Anthropic's format. Every other endpoint uses OpenAI's. Both carry the same code, which is stable, so match on it rather than on the message.

{
  "error": {
    "message": "Insufficient balance. Top up your account to continue.",
    "type": "insufficient_quota",
    "code": "insufficient_balance",
    "param": null
  }
}

Status codes

400 invalid_request: The request is malformed or uses a feature we don't support. The message names the parameter.
401 invalid_api_key: The key is missing, mistyped or revoked.
402 insufficient_balance: Your available balance can't cover the request. Add funds on the Billing page.
402 key_spend_limit_reached: The key has reached its spending limit. Raise or remove the limit on the API keys tab.
403 key_disabled, account_disabled: The key or the account has been disabled. Contact support.
404 model_not_found: The model ID doesn't exist or isn't available. Check it against GET /api/v1/models.
404 not_found: Unknown endpoint. Check the path and the base URL.
413 request_too_large: The request body is over 32 MB.
429 rate_limit_exceeded: Too many requests for the key, or too many at once for your account. Wait for the number of seconds in the Retry-After header.
502 upstream_error: The model provider returned an error or couldn't be reached. Retry with backoff.
503 service_unavailable: Requests are paused for a short time, for example during maintenance. Retry after the Retry-After delay.

Rate limits

Each key allows 120 requests per minute unless you give it its own limit, from 1 to 10,000, on the API keys tab. Going over returns 429 with a Retry-After header.

Concurrent requests

Your account can run a limited number of requests at the same time, across all keys. The limit grows with your available balance:

Less than $5: 4 requests at a time
$5 to $49.99: 16 requests at a time
$50 or more: 64 requests at a time

A request over the limit gets 429. It can be retried as soon as another request finishes.

Balance holds

While a request runs, it holds an estimate of its maximum cost from your balance. Your available balance is the balance minus these holds, and GET /api/v1/balance returns both.

A request fails with 402 insufficient_balance when the available balance can't cover its hold, even if the balance itself is higher. Holds also count toward a key's spending limit. Setting max_tokens keeps holds small.

Retries

Retry 429, 502 and 503 responses with exponential backoff, waiting at least as long as Retry-After when it's present. The official OpenAI and Anthropic SDKs do this for you.