The level of reasoning effort the model applies when generating a response. For supported values, see the model provider's documentation.