Abstract Phenomenological and physically based constitutive models were developed using true stress–true strain curves obtained through hot compressing test of Ti–5Al–4Mo–6V–2Nb–1Fe alloy. The experimental data acquired under the different temperatures (700, 800, 900, 1000 and 1100°C) and strain rates (0.001, 0.01, 0.1, 1 and 10 s–1) conditions were modified considering friction. Based on the modified experimental data, the modified Johnson–Cook model, Arrhenius type model (phenomenological models) and the microstructure–based model (a physically based model) were developed to predict the hot flow stress of the alloy. And for each model, the average absolute relative error and determination coefficient were evaluated by comparing the experimental data to the predicted stress value. The determination coefficients were 0.9937, 0.9949 and 0.9956 for the modified Johnson–Cook model, Arrhenius type model and microstructure–based model, respectively. And the average absolute relative errors were respectively 6.71, 5.59 and 3.57%. Through the comparison of the developed constitutive material models, it is found that the microstructure–based constitutive model possessed the highest prediction accuracy.