Fused filament fabrication (FFF) is an additive manufacturing process in which the materials like polymers and composites are extruded for layer by layer fabrication. In FFF process, the process planning consists of determination of part build orientation, support structure requirement, slicing and tool path planning. Many studies in literature have shown that choosing the correct build orientation in FFF is crucial for the improvement of the surface finish, mechanical strength of the part as well as to reduce build time. A reliable estimation of build time is vital as it is the basis for estimating production cost, process planning, and build orientation optimization (BOO). In the last two decades, many researchers have developed parametric methods to estimate build time for FFF processes that consider part volume and height, support volume, layer thickness to name a few. High-fidelity models for estimating build time use tool path simulation which consider many kinematic parameters of the machine like printing head acceleration, deceleration, printing speed etc. in determining accurately the build time. However, efficient applicability of these methods as multi-fidelity models in the context of BOO considering computational cost has not been well researched. This work thus initially evaluates the correlation coefficient of different build time estimate models with a high-fidelity model and the computational efficacy. Then the work proposes a hybrid optimization framework that uses multi-fidelity models to obtain optimum build orientation with improved computational performance. First, we do a multi-modal build orientation optimization by a Covariance Matrix Self-Adaptation Evolution Strategy with Repelling Subpopulations (RS-CMSA) algorithm using a low fidelity model and use these as the initial guess in a gradient-based optimization method using a high fidelity model. The proposed hybrid method has been illustrated with example case studies as well as compared and evaluated to a standard optimization algorithm using a single fidelity model to demonstrate the overall methodology and its effectiveness.​