I tried it on some code I found on github for processing video. It took 4 minutes per frame in pure python. With Numba I got it down to 20 seconds per frame so far. Mostly by just adding an @njit decorator on top of the functions, and separating loops into their own jitted functions. Unbelievable. It just needs better support for calls to third party non-numba libraries and classes. then I'm sold, I wouldn't use anything else. Right now if there are any complex classes or third party calls I can get stumped trying to 'Numbaify' it.