Software is the essential part for the users and the reliability of software provides its validity. To check the validity of the software, we have used the family of Lindley distribution such as Lindley distribution (LD), Beta-generalized Lindley distribution (BGL), and a new extension of Lindley distribution (NLD). Four different software failure datasets and real datasets is used for testing for the better fitment of all three distributions. Maximum likelihood method is used to estimate the parameters of related distributions. The performance of the models is based on their Akaike information criterion and Bayesian information criterion. The result revealed NLD superiority over the other two existing LD and BGL distributions for all software reliability datasets. The graphical representation provides a helpful illustration of how the implementation enhances the reliability forecast.