Mesh quality and turbulence modeling are two core influencing factors in supersonic numerical simulations, and their performance directly determines the accuracy and reliability of the simulations. This study systematically investigates turbulence modeling challenges in supersonic cavity flows, focusing on three k-ω turbulence models (k-ω TNT, k-ω BSL, and k-ω SST). Using two benchmark cases—the Settles slanted cavity and the MY standard combustor—the predictive accuracy of these models is comparatively evaluated, with an emphasis on analyzing their sensitivity differences to mesh scale and topological structure. Key findings include: k-ω SST Model exhibits exceptional stability across a wide range of mesh scales, maintaining excellent predictive performance under both coarse and fine mesh configurations. k-ω TNT Model demands strict mesh conformity, and is recommended for use with grids precisely matched to the flowfield’s topological structure. Such pairing significantly enhances the model’s ability to capture complex flow features. k-ω BSL Model shares similarities with k-ω SST in terms of insensitivity to mesh topology but demonstrates inferior predictive accuracy compared to k-ω SST. For complex flow structures (e.g., strong shear layers, shock-reflection interactions), if computational resources permit, the k-ω TNT model combined with high-fidelity meshes is prioritized to boost prediction precision. Conversely, in engineering scenarios with strict computational efficiency constraints or limited mesh quality, the k-ω SST model is the optimal choice due to its robustness and balanced performance, effectively mitigating risks of simulation divergence caused by mesh distortion.