The influence of turbulence modeling approach by means of (U)RANS and large eddy simulation (LES) on the overall modeling of turbulent condensing wet steam flows is investigated using the example of a low-pressure steam turbine cascade. For an accurate numerical treatment of turbulence in the presence of shock waves, necessary for predictive scale-resolving computations, a hybrid flux treatment switches between a baseline nondissipative central flux in energy consistent split form and a shock-capturing upwind flux in shocked regions based on a shock sensor. Condensation is realized by a monodispersed Euler–Euler source term model, the equation of state by the highly efficient and accurate spline based table lookup method (SBTL). The numerical treatment is validated with a decay of homogeneous isotropic turbulence test case containing eddy shocklets. The measurement results of the condensing wet steam cascade are overall much better matched by LES compared to RANS and URANS. Analysis shows that the LES is much better able to account for the inherently unsteady nature of the spontaneous condensation process and its interaction with the trailing edge shock wave structure.