DocsFeatures

OpenAPI specification

Model IDs and options in examples are illustrative. Select an available model from GET /v1/models and check its capabilities and current prices.

Reasoning effort

Set reasoning_effort on Chat Completions, or reasoning.effort on Responses, to none, minimal, low, medium, high, xhigh or max. Each provider of a model accepts its own levels, listed per offering in text.reasoning_control from GET /v1/models.

ParameterTypeDefaultDescription
supported_effortsstring[]—Levels this provider accepts, highest first. Empty when the model always reasons at a level you cannot change.
mandatoryboolean—true when reasoning cannot be turned off, so none is not accepted.
default_effortstring | null—The level used when the request sends none.
  • A listed level is sent as is.
  • An unlisted level uses the nearest listed one, and the lower one on a tie. Any level other than none stays on.
  • none on a model that always reasons skips that provider. When no provider is left, Ark returns 422 reasoning_required and calls no provider.
  • No level uses the provider’s default.
{
  "error": {
    "message": "This model always reasons on this provider. Set reasoning_effort to low or higher, or leave it out.",
    "type": "invalid_request_error",
    "param": "reasoning_effort",
    "code": "reasoning_required"
  }
}