The rapid growth of Internet of Things (IoT) devices has intensified cybersecurity risks due to their heterogeneous and resource-constrained nature. This paper proposes GTM-XAI, a Graph Transformer integrated with Explainable Artificial Intelligence (XAI), for accurate and interpretable IoT cyber threat detection. IoT communication patterns are modeled as dynamic graphs, and multi-head self-attention is applied to capture both local and global dependencies. To ensure transparency, SHapley Additive exPlanations (SHAP) values identify critical devices and communication links contributing to threat predictions, enabling targeted mitigation such as device isolation and link blocking. Experimental evaluation on Edge-IIoT, ToN_IoT, and CIC-IoT2023 datasets shows detection accuracies of 98.5%, 98.86%, and 98.67%, respectively, outperforming conventional baselines while providing interpretable outputs that support real-time incident response. The results demonstrate that GTM-XAI combines advanced detection capabilities with explainability, making it a practical solution for enhancing IoT cybersecurity.