Significance Accurately simulating high-Reynolds number turbulent flows is computationally challenging. Traditional turbulence modeling methodologies, e.g., large eddy simulation (LES), aim to resolve large-scale turbulence in the Eulerian setting, while modeling the subgrid contributions using the resolved-scale flow; the approach usually involves various physics assumptions and parameter tuning. In this work, we develop LES turbulence models, stated in terms of Lagrangian particles moving with the flow, using the physics-informed machine learning framework. We propose generalized equations to model the evolution of a Lagrangian particle cloud with physics-informed parameterization and functional freedom based on a representation using Neural Networks. We show that the proposed Lagrangian–LES model is capable of reproducing turbulence structure and statistics over a range of turbulent Mach numbers.