Computational methods to predict and tune electrochemical redox potentials are important for the development of energy technologies. Here, we benchmark several semiempirical models to compute reduction potentials of organic molecules, comparing approaches based on (1) energy differences between the S-0, and one-electron-reduced D-0 states of the isolated molecules and (2) an orbital energy shift approach based on tuning the charge-transfer triplet energy of the Ag-20-molecule complex; the second model enables explicit modeling of electrode-molecule interactions. For molecules in solution, the two models yield nearly identical results. Both PM7 and PM6 predict formal potentials with only a slight loss of accuracy compared to standard density functional theory models, and the results are robust across several choices of geometries and implicit solvent models. PM6 and PM7 show dramatically improved accuracy over older semiempirical Hamiltonians (MNDO, AM1, PM3, and INDO/S). However, our recently developed INDO parameters model the electronic properties of our Ag-20 model electrode much more accurately than does PM7. These results demonstrate the need for further development of semiempirical models to accurately model molecules on surfaces.