Handwritten text recognition (HTR) is the technique of recognizing and interpreting handwritten text into machine-readable output. HTR is a challenging problem given the variance in handwriting styles across people and the poor quality of the handwritten text. However, considerable work has been accomplished to recognize Latin scripts. In contrast, the accuracy of Arabic HTR systems is far behind the HTR of Latin script. In this paper, a comparative experimental assessment of four recent deep learning models (namely, FCN, GFN, VAN, and DAN) that have been proposed for HTR of Latin scripts. These models are evaluated on the KHATT dataset, a challenging Arabic handwritten text dataset. The lowest CER and WER are obtained using the DAN model. In addition, a deep analysis of the challenges related to the Arabic HTR is discussed.