<p>Optical Character Recognition (OCR) is a significant technological advancement that turns scanned documents and pictures with text into machine-readable formats. While OCR has reached high accuracy rates for Latin-based languages, such as English, there are still significant challenges for right-to-left languages, including Arabic. Improving the capabilities of Arabic Handwriting Recognition (AHR) is critical to enhancing access to digital content in the Arabic language, given that there are well over 400 million speakers. Arabic OCR faces unique challenges regarding the nature of the language, where letter forms change according to their position within a word with or without dots and connections between intricate characters. Moreover, diacritical marks and curved shapes present additional challenges. This paper presents a hybrid model for Arabic character recognition. It is based on Convolutional Neural Networks (CNNs) and various machine learning classifiers such as Support Vector Machines (SVM), k-nearest Neighbors (KNN), Random Forest (RF), Long Short-Term Memory (LSTM), and Kolmogorov Arnold Networks (KANs). The proposed CNN+KANs model exploits the ability of CNNs to extract features and the ability of KANs to approximate complex functions to achieve high recognition accuracy. The model is tested over two datasets, including diverse handwriting styles, such as the Arabic Handwritten Characters Dataset (AHCD) and Arabic letters written by children, which we call Hijja. Experimental results demonstrate that the CNN+KANs model achieves state-of-the-art performance, with testing accuracies of 97.71% on AHCD and 91.32% on Hijja, outperforming traditional models like CNN+SVM and CNN+KNN. The study shows that the CNN+KANs hybrid model is strong and can adjust well, which makes it a good choice for recognizing Arabic handwritten characters.</p>