Generally, steels with a yield strength more than 1400 MPa are defined as ultrahigh-strength steels. [1] Ultrahigh-strength steels own exceptional mechanical qualities and are utilized to create large key load-bearing components, [2][3][4] which are commonly treated by hot forging with considerable strain. During the hot forging process with large strain, the mechanisms of dynamic recovery and dynamic recrystallization occur, which make it difficult to control the flow behaviors. [5,6] In addition, flow behaviors closely depend on strain, strain rate, and deformation temperature during the hot deformation process. The flow characteristics are dramatically different at various deformation parameters. The numerous features make it challenging to forecast flow characteristics and optimize hot processing parameters. [7] Constitutive modeling is a powerful method to simulate the real deformation process. [8] So, it is vital to analyze flow characteristics and develop an appropriate constitutive model for ultrahigh-strength steels during the hot forging process with large strain.Constitutive models include artificial neural network models, physical-based constitutive models, and phenomenological constitutive models. [9,10] Artificial neural network models can accurately capture the intricate link between flow characteristics and multiple deformation factors. Wan et al. [11] utilized the methods of back propagation neural network and particle swarm optimization to predict the flow behaviors of Zr-4 alloy during hot deformation, and the correlation coefficient between predicted and experimental flow stress is 0.9981. Li et al. [12] employed a deep neural network to characterize the flow characteristics of Ti-2Al-9.2Mo-2Fe beta titanium alloy during hot deformation. The prediction accuracy of the constitutive model is 0.9995. Murugesan et al. [13] explored the accuracy of different supervised machine learning methods in predicting flow stress and pointed out that the random forest regression model has the highest accuracy. The flow behaviors can be correctly predicted by artificial neural network models. However, as stated by Savaedi et al., [14] the successful application of artificial neural network models depends on high-quality data. Furthermore, compared to the computation efficiency of analytical models, the computation efficiency of artificial neural network models is lower. [15] Physical-based constitutive models may express the physical meanings of material during the deformation process. Buzolin et al. [16] constructed a dislocation-based constitutive model to predict the flow behaviors and microstructure evolution during the hot deformation process of Ti5553 alloy. Zeng et al. [17] built a unified constitutive model