With the advancement of technology today, computer software has a high effect in different areas of our life. In particular open-source software is used by many software companies because it helps them to create their new software without implementing it from scratch. Therefore, the quality of open-source software is a significant issue and one of the most popular research topics in the literature. Checking software reliability during the development cycle is an indicator of releasing the software or not. In this paper, we propose to use Marine Predators Algorithm with six probabilistic models for assessing reliability more accurately. We select three versions of a standard dataset for GNU's Not Unix Network Object Model Environment projects. We compare the implemented models by three evaluation criteria: mean square error, sum of square error, and reliability. The results for all versions of the dataset show that SRGM-5 is based on the methodology of imperfect debugging estimates the most accurate reliability results in terms of mean square error and sum of square error.