Great question. So it's different in a few ways. Calling a tool on a reasoning model is a pretty undeterministic thing, meaning that (of course you could control the max thinking tokens, but you still don’t know when it’s gonna stop reasoning). This gives much more control,