It is possible to use generic Python objects as the dtype for a NumPy array, but if you do this, you’ll get no better performance with NumPy than you would with Python generally.