The hybrid static synchronous compensator (hybrid-STATCOM) is characterized by a wide compensation range and low dc-link voltage, which is a cost-effective reactive power compensator for medium voltage level application. However, the coupling part of the hybrid-STATCOM is time-varying, and its system model is nonlinear, which causes a great challenge for the controller design. This letter proposes a model-free reinforcement learning (RL) based sliding mode control (RL-SMC), which provides the inverter voltage as the control action of the hybrid-STATCOM to compensate for the load reactive power and harmonic. The proposed RL-SMC is computationally efficient with high steady-state accuracy, fast response, and good robustness. First, an agent-environment framework is proposed to enable RL. Then, the comprehensive design procedure of the RL-SMC is proposed. Finally, simulation and experimental results are carried out to verify the validity and effectiveness of the proposed RL-SMC under different load and grid conditions.