This is very true. I'm more careful with what I ask my local model, since it's about 1/3 as fast as the cloud models I'm used to.