Abstract: Fiber reinforced polymer (FRP) is widely used in enhancing the performance of concrete structures and strengthening damaged components due to its advantages of light weight, high strength, corrosion resistance and convenient construction. The ultimate conditions of FRP-confined concrete are the important factors that must be considered in the selection of FRP types, FRP thickness and the number of covering layer. The prediction results of the existing ultimate stress model can better reflect the real situation, while the prediction accuracy of the existing ultimate axial strain model is low, so the ultimate axial strain was studied. Since there are many factors that affect the ultimate axial strain of FRP-confined concrete, the models proposed by many researchers have large differences in the choice of input parameters. Therefore, the influence of different input forms on the prediction accuracy of ultimate axial strain model was discussed while the ultimate axial strain model was established by gene expression programming. Five statistical indicators such as coefficient of determination and mean absolute error were used to evaluate the prediction results of model, which was compared with the existing prediction models. The research results show that the model corresponding to the input form of the combination of original data and new data has the highest prediction accuracy, so the selection of model input parameters should not only consider the original data or new data. Compared with the models proposed by other researchers, the prediction accuracy of the model proposed in this article is the highest. The coefficient of determination is 0.893, and the mean absolute error and other indicators are all below 0.35.