Limits

{
  "error": {
    "message": "Rate limit: 60 requests per minute per key."
  }
}

# Limits

A request over a limit gets a 400 or a 413 that says which.

Built and tested, not yet deployed. Early access.

## Limits

| Limit | Value |
| --- | --- |
| Request body | 256 KB. Checked on `Content-Length`, and again on the body’s length once read. |
| Your question’s `instructions` | 1 to 4,000 characters |
| Your question’s text `input` | Up to 20,000 characters |
| `options` | 2 to 50 |
| `levels` | 2 to 11 |
| `abstainBelow` | 0.5 to 0.99 |
| Merchant and descriptor text | 1 to 200 characters |
| Transaction `id` | 1 to 100 characters |
| Amount | Up to 1,000,000,000 US dollars either way |
| `transactions` | Up to 500, ids unique |
| Candidates the model sees | 30: the nearest in time for receipts, the earliest for refunds |
| Currency | US dollars only |
| Active keys per account | 10 |
| Batch | 1 to 100 requests, in a body of up to 1 MB |
| `Idempotency-Key` | 1 to 255 printable characters, kept 24 hours |

## Rate limit

60 requests per minute per API key or connected app, counted over the past minute. The next call gets a 429 with `Retry-After: 10`. Answered calls, and calls refused for their fields or for sensitive data, count toward it; a replayed `Idempotency-Key` doesn’t.

Response: 429 Too Many Requests

```json
{
  "error": {
    "message": "Rate limit: 60 requests per minute per key."
  }
}
```

## Credit

A call with an API key or a connected app’s token needs credit on the account, checked before the call. At zero, every call gets a 402 until credit is added. A call that fails costs nothing. The planned price and the free credit are on [Pricing](/api/docs/pricing).

Source: /api/docs/limits