Instead of calling a huge, math-heavy Python/NumPy simulation model on every iteration JET packs the whole model in a single performant C++ function which results in an enormous speed-up (up to 60-fold speed up was observed in a simulation of a math-heavy quadcopter-model).
from jet.jit import jit import scipy.integrate from simulation_model import state_derivatives, init_state, time_vec jet_state_derivatives = jit(state_derivatives) states = scipy.integrate.odeint( jet_state_derivatives, init_state, time_vec)