This study aims to price ELS options using the DeepBSDE model, a novel deep learning approach for solving backward stochastic differential equations. ELS options present significant challenges due to their intricate payoff structure and path-dependent nature. We experimented with two types of ELS options: stepdown knock-in and knockout options. The DeepBSDE model shows similar results to those of classical methods like Monte Carlo method and Finite Difference method, but with distinct advantages. While Monte Carlo method provides only the option's value without any additional information such as the value of Greeks, and while Finite Difference method offers comprehensive domain information but struggles with multi-dimensional problems, DeepBSDE avoids the curse of dimensionality, making it more practical for usage in multi-asset options. The results of our experiment reveal that despite its initial longer training times, the DeepBSDE model’s inference time is comparable to that of the classical methods, hence minimizing computational burden once the model is trained. Such efficiency makes the DeepBSDE model particularly suitable for markets in countries like South Korea, where stepdown ELS options dominate, constituting 70% of the market transaction. The DeepBSDE model’s flexibility also allows for future extensions to evaluate ELS options with multiple underlying assets and incorporate correlations between such assets. Additionally, enhancing the model to consider multiple early redemption points will further improve its practical application.