Early detection of eye diseases, such as Diabetic Retinopathy (DR), Cataract, and Glaucoma, as well as the ability to accurately differentiate these from Normal conditions, is critical for preventing irreversible vision loss and improving patient outcomes. However, accurate and automated classification from fundus images remains a challenge due to the scarcity of labeled data and the limitations of conventional supervised learning. To address this, we propose a novel hybrid learning framework that integrates Momentum Bootstrap Your Own Latent (MoBY) with a Swin Transformer backbone for precise and robust multi-class eye disease classification. The proposed model first undergoes self-supervised pretraining to extract discriminative features from unlabeled images, followed by supervised fine-tuning on limited labeled data for task-specific adaptation. HyMoBY-Swin employs the Swin Transformer backbone for deep feature extraction. These features are refined through contrastive learning by maximizing similarity between augmented views of the same image while minimizing overlap with unrelated samples. Supervised fine-tuning further enhances generalization across both labeled and unlabeled data. Extensive experiments on the OIH dataset, comprising 4,215 fundus images, demonstrate that HyMoBY-Swin achieves a robust test accuracy of 92.43%, with consistently strong performance across all four categories. Diabetic retinopathy demonstrated the highest precision and sensitivity, both at 0.98, along with a leading ROC-AUC score of 0.99, followed by cataract (0.96), glaucoma (0.93), and normal cases (0.92). Ablation studies reveal that HyMoBY-Swin significantly outperforms its individual components and other state-of-the-art baselines, demonstrating the effectiveness of the combined contrastive learning and transformer-based architecture. Furthermore, t-SNE visualizations confirm the model’s ability to clearly differentiate between disease categories, highlighting its strong generalization capacity across complex clinical scenarios. These results establish HyMoBY-Swin as an accurate and generalizable solution for real-world ophthalmic settings, offering significant clinical utility by reducing reliance on large annotated datasets and enhancing diagnostic efficiency.