In the present work, the prediction of flow stress for IN625 has been done using two of the latest constitutive models, namely, modified Arrhenius (m-A) and combined Johnson-Cook and Zerilli-Armstrong (JC-ZA), and also by an artificial neural network (ANN) at 300, 473, 673, 873, and 973 K. On comparing the predictive efficiency of flow stress using constitutive models, the m-A model displayed better statistical parameters. It includes the effect of activation energy, temperature, and strain rate while predicting the flow stress, which are the most critical parameters while working at elevated temperature. The ANN approach helped predict the best flow stress results with the least average absolute error (AAE) of 0.93 % and the highest correlation coefficient (R) of 0.992. Further, the Barlat 1989 yielding function best predicted the yield loci among the Hill 1948 and Barlat 1989 criteria. The experimental stretch forming test has been conducted, and the forming limits improved by approximately 30.25 % as the temperature increased from 300 to 973 K. Additionally, the theoretical model, namely, Marciniak–Kuczyński (MK) model, has been used by inducing the combined effect of the constitutive model and yielding function for predicting the forming limits. The m-A constitutive model, in combination with Barlat 1989 yielding function, best predicted the forming limit curve. The ANN method has also been used after rigorously training the neural network, and it displayed the least AAE while predicting the forming limits. Thus, the neural network technique can be used as a modern tool to predict results involving extreme virtual experimental conditions.