Reynolds-averaged Navier–Stokes (RANS) model-based conjugate heat transfer (CHT) method is so far popularly used in simulations and designs of internally cooled gas turbine blades. One of the important factors influencing the RANS-based CHT method's prediction accuracy is the choice of turbulence models for different fluid regions because the blade passage flow and internal cooling have considerably different flow features. However, most studies in the open literature adopted the same turbulence models in the blade passage flow and internal cooling. Another important issue is the comprehensive evaluation of the losses caused by the flow and heat transfer for both fluid and solid regions. In this study, a RANS-based CHT solver suitable for subsonic/transonic flows was developed based on OpenFOAM and then validated and used to explore suitable RANS turbulence model combinations for internally cooled gas turbine blades. Entropy generation, being able to weigh the losses caused by both flow friction and heat transfer, was used in the analyses of two vanes with smooth and ribbed cooling ducts to reveal the loss mechanisms. Findings indicate that the combination of the k–ω SST–γ–Reθ transition model for passage flow and the standard k–ε model for internal cooling provided the best agreement with measurement data. The relative error of vane surface dimensionless temperature was less than 3%. The variations of entropy generation with different internal cooling inlet velocities and temperatures indicate that reducing entropy generation was contradictory with enhancing heat transfer performance. This study, which provides a reliable computing tool and a comprehensive performance parameter, has an important application value for the design of advanced internally cooled gas turbine blades.