> especially as I presumed NumPy to do far better If you are running your benchmark on a single Numpy operation then I would expect Numpy to have equal or bett

> especially as I presumed NumPy to do far better If you are running your benchmark on a single Numpy operation then I would expect Numpy to have equal or better speed (you are not paying for JIT compilation). However, when you are doing several operations, Numpy will do a loop on array elements for each operation while JAX can fuse everything, that can end up making a big difference.