The growing global burden of vision-threatening eye diseases like glaucoma, diabetic retinopathy, and cataracts demands more scalable diagnostic solutions. We present YOLO-ViT, an innovative hybrid model that synergizes YOLOv8’s localized feature extraction with Vision Transformers’ global attention mechanisms for automated ocular pathology detection. Our model uniquely combines the strengths of convolutional architectures and transformer models, achieving state-of-the-art classification performance while requiring significantly fewer training epochs than conventional CNN-based methods. Ex-tensive evaluations demonstrate the model’s superior diagnostic capabilities, particularly in challenging glaucoma detection scenarios where existing approaches often falter. The proposed solution offers clinicians an accurate, efficient, and scalable alternative to manual screening, with promising potential for deployment in resource-constrained healthcare environments. Compared to conventional CNN architectures, the proposed YOLO-ViT model demonstrates superior performance, achieving accuracy and F1-score of 91.93% (0.92) at 20 epochs, 94.48% (0.94) at 30 epochs, and 96.89% (0.97) at 40 epochs. In contrast, EfficientNet B3, Inception V3, and VGG-19 report lower results: 93.00% (0.93) at 40 epochs, 89.00% (0.88) at 50 epochs, and 88.00% (0.87) at 60 epochs, respectively highlighting the effectiveness and efficiency of the YOLO-ViT model.