This talk presents a new modelling and control paradigm for distributed large-scale coordination of “smart” thermostatic loads. These appliances are assumed to operate independently in response to price signals, scheduling their power consumption and allocating frequency response provision to minimize their individual cost. A mean field game framework is used to characterize the behaviour of the single loads and quantify their overall impact on the power system. This is combined with the resolution of a unit commitment problem, which returns meaningful price signals while accounting for different generation technologies and the security constraints of the grid. Ad hoc anti-synchronization techniques are introduced to ensure convergence to an equilibrium solution and simulation results are presented to quantify the potential value of flexible thermostatic loads in the decarbonized power system of the future.