Floating-point operations performed during model inference; a measure of computational cost independent of hardware speed.