Abstract—The Hadiths are a compilation of the sayings, actions, and approvals of the Prophet Muhammad (PBUH). They serve as guidelines for the lives of Muslims after the Quran. With thousands of hadiths, determining their authen- ticity based on grades of sahih (authentic), hasan (good), and daif (weak) using deep learning is necessary. Deep learning (DL) has the potential to markedly enhance the accuracy of hadith classification by capturing intricate text patterns and automating the classification process, which is beneficial for handling large datasets such as hadiths. In this study, we used BERT-BiGRU and BERT-BiLSTM models to determine the authenticity of hadiths based on their grades. First, the dataset was preprocessed with punctuation removal, stop words removal, and stemming. Then, one-hot encoding was applied for the hadith grade categories, followed by oversampling to balance the dataset. After that, the processed dataset was used with BERT-BiLSTM and BERT-BiGRU models. The results of this research indicate that the BERT-BiLSTM model demonstrates superior performance compared to the BERT-BiGRU model, achieving an accuracy of 0.963, precision of 0.965, recall of 0.963, and F1-score of 0.963.