Sequential Approximate Optimization (SAO) is a general framework for finding the optimum of a problem without the need to perform a large number of evaluations or to know function gradients beforehand [1]. It is an interesting alternative for problems that require expensive computational simulation to evaluate a single response. The framework can be summarized by three different steps. First, an initial sampling design is defined, where High-Fidelity (HF) analyzes will be drawn. Second, a metamodel is fitted to the data, which will be used to assist in the optimization of the High-Fidelity problem. Third, a new sampling point is chosen based on the metamodel response and an infill criterion. The second and third steps are performed iteratively until a stopping criterion is reached [2]. A Multi-Fidelity (MF) metamodel is a kind of surrogate model that allows for the consideration of information from a lower fidelity function, alongside information from the high-fidelity function [3]. An example would be using a fine finite element mesh as the HF model and a coarse mesh as a low-fidelity (LF) model. The MF approach allows for the use of a larger sampling space at a much lower cost to build the surrogate model. The choice of new sampling points can also be performed by considering these multiple fidelities, allowing for a more efficient identification of the optimal space [3]. In this work, we aim to perform SAO with MF models to reduce the computational cost of the optimization of Functionally Graded Plates (FGPs) composed of metal-ceramic composites. Our goal is to find the optimal material gradation of the FGP while subject to a constraint related to the maximum total volume fraction of one of the constituents. Problems related to the maximization of the fundamental frequency and maximization of the buckling load, considering FGPs with different materials, geometries, and boundary conditions, are solved. Furthermore, both unidirectional and tridirectional material gradations are considered. The HF and LF evaluations are performed using Isogeometric Analysis (IGA), considering 3D solid and High-order Shear Deformation Theory (HSDT) plate models. Computational results are obtained using the software BIOS [4]. The numerical results show a significant decrease in the computational cost provided by the MF approach with respect to the optimization using single fidelity surrogate models.