Kind of. Those that are explicitly trained to do that with consistent formats will do it better. They'll also save you the extra tokens needed to explain the format/method of interacting with functions. But yeah, you can simulate this with any recent model and enough explanation.