There is context caching in many models. It is less expensive if you enable that.