But even that isn't the whole story because the models can produce wildly amount of thinking output as well as regular output for a similar query. Sometimes you

But even that isn't the whole story because the models can produce wildly amount of thinking output as well as regular output for a similar query. Sometimes you can take a cheap model and have it think a ton or an expensive model that thinks little and get similar results. But the number of tokens generated will be wildly different.