Laser powder bed fusion (LPBF) additive manufacturing of metals offers a unique potential for the production industry with near-unlimited freedom of design. The underlying physical mechanisms of this relatively young technology and their interplay are not yet fully understood, but this understanding is crucial to predicting optimal process parameters for producing high-quality parts. In particular, the highly dynamic melt pool thermo-hydrodynamics, accompanied by rapid evaporation, may entail quality-degrading defects, such as evaporation-induced pores, spatter, and lack of fusion under typical process conditions. State-of-the-art melt pool models often rely on a diffuse representation of the metal-gas interface, which is subject to extreme interface fluxes and forces and undergoes complex topology changes [1,2]. While diffuse interface models provide a mathematically consistent formulation for two-phase flow problems involving rapid evaporation [3], our recent contribution demonstrates that the extreme temperature gradients combined with the high ratios of material properties between metal and ambient gas lead to significant errors in the interface temperature when typical interface thicknesses and discretizations are applied [4]. The magnitude of evaporation-induced fluxes and forces depends exponentially on the interface temperature [5], which, therefore, must be predicted with high accuracy. The present work presents a novel high-fidelity modeling approach for the melt pool dynamics in LPBF. Employing a sharp interface CutFEM approach for the multi-phase thermal problem, the interface temperature is predicted with high accuracy, allowing accurate temperature-dependent interface forces for the diffuse interface multi-phase flow model to be computed. By coupling a sharp interface thermal model with a diffuse interface flow model, the strengths of both approaches can be exploited, namely an accurate interface temperature prediction and an efficient and robust multi-phase flow solution scheme.