Rate Limits

The public v1 API protects queue capacity, worker capacity, and input size. Size limits are part of the OpenAPI schema and are enforced before jobs are accepted.

Score limits:

up to 100 items per job
up to 20,000 characters per item
up to 100,000 total characters per job

Rewrite limits:

up to 10 items per job
up to 5,000 characters per item
up to 20,000 total characters per job

Pagination limits:

limit defaults to 50
limit must be between 1 and 100

Operational request-rate and queue-concurrency limits may be adjusted to protect GPU capacity. Clients should use idempotency keys on create requests, poll with backoff, and retry only retryable failures. The current public OpenAPI contract does not publish a fixed numeric requests-per-minute quota.

When an operational limit is exceeded, create requests return 429 with error.code set to rate_limited. Requests denied by billing policy return 402 with error.code set to billing_disabled. These failures happen before job creation, artifact writes, callback token generation, or task dispatch.