This paper introduces ViTDroid, a novel hybrid model that combines Vision Transformers (ViTs) and recurrent neural networks (RNNs) to enhance Android and IoT malware detection. ViTDroid addresses critical challenges by leveraging ViTs to capture global spatial dependencies and RNNs (LSTM and GRU) to model temporal patterns, enabling comprehensive analysis of complex malware behaviors. Additionally, the model integrates explainability tools, such as LIME and SHAP, to enhance transparency and trustworthiness, essential for real-world cybersecurity applications. The study evaluates ViTDroid's performance against conventional models, including RNN, LSTM, and GRU, using accuracy, precision, recall, and F1 score as evaluation metrics. Results demonstrate that ViTDroid achieves superior performance with an accuracy of 99.1% for Android malware and 98% for IoT malware. Precision and recall values reach 0.99 and 0.98, respectively, for Android, and 0.97 and 0.98 for IoT, with F1 scores of 0.99 for Android and 0.97 for IoT. These findings underscore ViTDroid's potential as a robust, efficient, and explainable solution to combat evolving threats in mobile and IoT ecosystems, paving the way for future advancements in malware detection systems.