In this work, the capabilities of state-of-the-art turbulence models are compared for a three-dimensional flow (3D) field within a constricted vertical pipe. The considered flow domain is a vertical pipe section with a baffle-shaped flow constriction which leads to the development of a jet flow through and a recirculation flow region behind the constriction. Different Reynolds-Averaged Navier-Stokes (RANS) and Large Eddy Simulation (LES) models were tested for single- and two-phase flow simulations. In the two-phase simulations, bubble-induced turbulence (BIT) was also considered by adding source terms in the k and ε/ω equations. The results are validated against experimental data. We employed hot-film anemometry (HFA) for liquid velocity measurement and combined it with ultrafast X-ray computed tomography (UFXCT), which provides gas phase data. Based on the local phase-indicator function obtained from the tomographic image data, we can correct HFA signals, which become corrupted by bubble contacts. We found that for single-phase flow all RANS models predict axial velocity well while radial velocity prediction is inadequate. LES models, however, achieve a better prediction of the latter. For two-phase flow, the axial component of the liquid velocity is well captured by all RANS models and the radial component of the liquid velocity is predicted better than for single-phase flow. In general, the computationally less costly RNG k-ε model performs similar to the SSG RSM model and can therefore be recommended for simulation of complex flow scenarios.