In order to embody the basic characteristics of flexible joint hysteresis behavior, a hysteresis-like operator was constructed. By connecting the hysteresislike operator and online sequential extreme learning machine (OS-ELM) in series, an OS-ELM hysteretic hybrid model of the flexible joint hysteresis characteristics for industrial robots was designed. In the hysteretic hybrid model, using the OS-ELM with high learning efficiency and strong generalization ability, the problems of the slow training speed and local minimum in learning the model parameters by using the gradient descent method could effectively avoid, and the modeling accuracy was improved. The model was validated by using the experimental data under the different input conditions. The results show that the proposed hysteretic hybrid model has high accuracy and higher generalization ability