Oral cancer is a critical health concern, accounting for the majority of malignancies in the oral cavity, with tongue cancer being the most prevalent in India. Early detection significantly improves survival rates, yet diagnostic delays remain a concern. This study explores the application of deep learning models in oral cancer detection. Leveraging the Oral Cancer Image (OCI) dataset, comprising diverse multimodal images, the research evaluates the efficiency of baseline CNNs, EfficientNet, and ViT architectures. Among these, EfficientNet emerged as the most effective model, achieving superior accuracy and precision compared to traditional CNNs and ViTs. To overcome dataset-specific challenges, we implemented data preprocessing and data augmentation techniques, and optimized model configurations, demonstrating the potential of hybrid deep learning frameworks. These findings highlight the impact of integrating advanced AI-models into clinical workflows, providing a foundation for precision medicine approaches in oral cancer diagnostics and contributing to the broader field of AI-driven healthcare solutions.