Thermoplastic vulcanizates (TPV) possess highly complex microstructures and heterogeneous phase compositions, making it difficult for traditional constitutive models to accurately capture their stress–strain relationships. This limitation significantly hinders the effectiveness of conventional mechanical modeling and predictive analysis. To overcome this challenge, the present study proposed a novel modeling framework that has integrated deep learning with finite element analysis. Within this framework, deep learning models was embedded into the Gaussian integration points of the finite element mesh to predict stress directly from strain, effectively replacing complex constitutive equations and enabling the efficient simulation of TPV mechanical behavior. Recognizing the history-dependent nature of TPV under loading, GRU model was employed for training purposes. The feasibility of the proposed framework was validated through simulations of uniaxial tension and compression. Based on the simulation results, the stress–strain evolution of TPV under various loading conditions was analyzed. Compared with conventional finite element methods, the proposed approach can accurately and efficiently capture the stress responses of TPV. This research provides a novel method for understanding and predicting the mechanical properties of TPV materials.