Breast cancer is an increasingly significant health issue, especially in urban areas of India. The likelihood of improved survival rates is closely linked to the effectiveness of early detection. This research employs an advanced Vision Transformer (ViT) model to analyze mammography images for the automatic diagnosis of breast cancer. A dataset comprising 13,128 images was employed, of which 5,970 were classified as benign and 7,158 as malignant. This collection underwent preprocessing techniques such as data augmentation, normalization, and resizing to enhance the effectiveness of the training process. To enable binary classification, a transfer learning approach was applied to adapt the ViT base model. In line with standard data distribution practices, 70% of the dataset was designated for training, 15% for validation, and the final 15% for testing. The optimization procedure utilized CrossEntropyLoss in conjunction with the AdamW optimizer. To evaluate its effectiveness, our Vision Transformer (ViT) model was compared to conventional CNN-based models, including ResNet and various VGG architectures. Furthermore, a web application built on Streamlit was created to facilitate real-time image classification, thereby enhancing accessibility for users without technical expertise.