The accuracy of post-B3LYP functionals is analyzed using an open-shell database of Cu(II) dinuclear complexes with well-defined experimental values of the magnetic coupling constants. This database provides a sound open-shell training set to be used to improve the fitting schemes in defining new functionals or when reparametrizing the existing ones. For a large set of representative hybrid exchange-correlation functionals, it is shown that the overall description of moderate-to-strong antiferromagnetic interactions is significantly more accurate than the description of ferromagnetic or weakly antiferromagnetic interactions. In the case of global hybrids, the most reliable ones have 25-40% Fock exchange with SOGGA and PBEO being the most reliable and M06 the exception. For range-corrected hybrids, the long-range corrected CAM-B3LYP and omega B97XD provide acceptable results, and M11 is comparable but more erratic. It is concluded that the reliability of the calculated values is system and range-dependent, and this fact introduces a serious warning on the blind use of a single functional to predict magnetic coupling constants. Hence, to extract acceptable magnetostructural correlations, a "standardization" of the method to be used is advised to choose the optimal functional.