We investigated the binding of late first row transition metals with chalcogen-chalcogen bridges represented by minimal models (H2O2, H2S2, and H2Se2). The use of such small models allows us to employ a large atomic basis set and compare DFT and MP2 results with CCSD(T) reference data. All methods agree in finding Cu2+ complexes the most stable ones, and for each given metal, H2Se2 complexes are more stable than H2S2 ones and the latter more stable than the corresponding H2O2 ones. Despite this qualitative agreement between all the considered methods, quantitatively we found a big difference between MP2 and B3LYP, in structural and energetic properties. In particular, DFT largely overestimates the binding energies, while MP2 slightly underestimates them with respect to CCSD(T) calculations. Note that also other popular functionals (MPW1PW91, M05, TPSS, BLYP, and SVWN) overestimate the binding energy, such that it seems to be an intrinsic DFT failure. The main discrepancy was found for Cu2+. The comparative analysis of B3LYP and MP2 wave functions explains the differences found between two methods and why the Cu2+ complexes show the bigger one. Finally, CCSD(T) calculations, slightly modifying MP2 insights, found that all three complexes present the same metal binding energy order, and notably Cu2+ > Ni2+ > Zn2+ > Co2+ > Fe2+ > Mn2+.