Enable separate thinking output in addition to content.
When true, the model returns separate thinking output alongside regular content. For supported models, this can also be set to a string ("high", "medium", "low") to control the level of thinking detail.