Solving problems in aerodynamics often relies on costly experiments or computationally intensive simulations. For example, the design optimization of aircraft wings typically involves computational fluid dynamics simulations for various shape parameters. However, performing several hundreds or even thousands of simulations, where each of them takes several hours or days, is simply infeasible. This issue is tackled by surrogate models, which constitute simplified mathematical models capable of approximating high-fidelity simulations at lower computational costs. A particular approach to surrogate modeling is given by neural operators, which represent deep learning architectures that directly approximate the solution operators to the partial differential equations underlying the physical problems. Thereby, they offer the possibility to solve entire classes of problems parameterized e.g. by different boundary or initial conditions. In this presentation, we compare the architectures as well as the numerical performances of multiple state of the art neural operators, such as Transolver, Geometry Aware Operator Transformer (GAOT), Universal Physics Transformer (UPT) and Bi-Stride Multi-Scale Graph Neural Network (BSMS-GNN). We evaluate the models of interest on two benchmark cases: First, we assess the models' performances in the prediction of the surface pressure distribution on a dataset of various airfoil shapes, comprising typical challenges in aerodynamical problems such as discontinuities (shocks) in the solutions. Second, we asses their performances in the same task on a dataset of different parameterizations of an aircraft configuration, evaluating further the models' abilities to handle complex geometries and to scale to industrial datasets.