Six years ago when I was working with a Phoenix API, we were measuring responses in microseconds on local dev machines, and under 5 ms in production with zero optimization. In comparison the identical Django app had a 50 ms floor.