Topic modeling is a crucial tool for understanding textual data, but its application in Arabic remains limited. This article addresses this gap by presenting an innovative study focused on Arabic topic modeling. We individually evaluate the performance of BERTopic, Non-negative Matrix Factorization (NMF), Latent Semantic Analysis (LSA), Latent Dirichlet Allocation (LDA), and leading Arabic language models (AraBERT, AraELECTRA, CAMeLBERT, QARIB, ALBERT). By analyzing these methods separately and in combination, we aim to ascertain the most effective approach for Arabic topic modeling. Our approach was tested on the Dataset for Arabic Classification using the Topic Coherence metric and showed promising results compared with the currently available approaches, achieving a coherence as high as 56% when using the AraBERT model.