Target localization utilizing frequency diversity array (FDA) is one of the most popular research directions in the study of radars. Considering the problems of range ambiguity and localization accuracy, this paper investigates the issue of three-dimensional (3D) target localization based on the proposed frequency diverse-subaperturing multiple-input multiple-output (FDS-MIMO) radar. First, a fifth-order tensor signal model is established by exploiting the inherent multidimensional structure of the matched filter output. To alleviate the difficulty of range ambiguity, the idea of employing co-prime frequency offsets along the two directions in the planer array is provided. This strategy can resolve the contradiction between the range resolution and maximum unambiguous range in beam domain. The tensor-based complex- and real-valued rotational invariance technique (tensor- and unitary tensor-ESPRIT) algorithms are developed based on frequency offset structure. Given that the inherent multidimensional structure is utilized, the proposed methods can resolve range ambiguity with improved localization performance as compared with the existing ESPRIT algorithm and frequency offset manner. Theoretical analysis and numerical results demonstrate the effectiveness of the proposed approaches.