Simple testing shows it does not use the naive sum. >>> import numpy as np >>> arr = np.array( (1_000_000_000.1, 1.1) * 50_000) >>> arr.mean() 50000000

Simple testing shows it does not use the naive sum. >>> import numpy as np >>> arr = np.array( (1_000_000_000.1, 1.1) * 50_000) >>> arr.mean() 500000000.6000003