The strength of “weak” models