The early stages of wave farm design require many parametric studies, e.g., regarding geometrical optimization of the array layout. Hence, mid-fidelity numerical models are employed due to their computational efficiency. In this study, the accuracy of these models and the necessary quality of input data (e.g., from Boundary Element Method simulations) is investigated using different calibration approaches. A heaving point absorber array comprising 24 devices, which are connected using a rigid frame, is used as the example wave farm. Three different approaches of model calibration are compared: (i) a low-effort approach without any calibration of the input data; (ii) an approach with medium-effort calibration based on experimental data of a single point absorber; and (iii) an approach with high-effort calibration based on experimental data of a whole wave farm. After a comparison with experimental data, the wave farm's power output using the three approaches is calculated and the accuracy as well as the implications for further design stages are discussed. The mid-fidelity hydrodynamic model can reproduce the mechanical interactions in the wave farm accurately, while the medium effort calibration shows high applicability due to the strong influence of the single point absorber calibration on the wave farm's power output.