The dynamics of underwater vehicle-manipulator systems (UVMSs) are very hard to model, which reduces the feasibility of model-based control approaches. Even so, such strategies prove useful in redundancy resolution. In this paper, higher-order sliding mode control is combined with task-priority operational space control (OSC) in order to handle and utilize the redundancy of UVMSs despite the presence of dynamic model errors. At each task level, the generalized super-twisting algorithm is implemented to reject effects caused by model errors while maintaining a continuous control signal. The general problem of OSC with uncertain models is analyzed, and some of its challenges are highlighted, including an algebraic loop. We conduct a simulation study on a highly redundant UVMS, where we compare task-level higher-order sliding mode control to proportional-derivative control. Though this paper is motivated by challenges specific to UVMSs, the results also hold for other vehicle-manipulator systems.