<p>Because of the intricate relationships between alloy composition and service conditions, it is still difficult to predict the mechanical properties (MP) of austenitic stainless steels (ASS), which are crucial for high-temperature applications. In order to simulate the history-dependent mechanical behaviour of austenitic stainless steels, this work proposed Inelastic Constitutive Artificial Neural Networks (ICANN), a physics-inspired neural network that incorporates internal state evolution. The main tensile properties of ASS that are predicted using ICANN are ultimate tensile strength (UTS) and yield strength (YS). To train and test the proposed model, 72 experimental entries from the National Institute for Materials Science (NIMS, Japan) are used. The ICANN method is implemented in MATLAB and it achieved high predictive performance for YS and UTS with a root mean squared error (RMSE) of 7.5&#xa0;MPa across five-fold cross-validation. The proposed model outperformed Generative Adversarial Network (GAN), Artificial Neural Network (ANN), and Deep Neural Network (DNN) models with higher a coefficient of determination (R²) and lower Mean absolute percentage error (MAPE). Sensitivity analysis and validation using different datasets are conducted to evaluate the model’s generalization. These results show that the ICANN model offers interpretable insights on composition property relationships in addition to improving predictive performance.</p>