Rheumatoid arthritis (RA) is an instance of arthritis that causes inflammation in the hands, legs, neck, and wrist joints. Signs of RA include joint pain, swelling across multiple joints, weariness, anorexia, and trouble moving in the morning. Early and accurate diagnosis of RA is crucial to manage symptoms and prevent irreversible joint damage. Recently, convolutional neural network (CNN) transformers have gained a crucial role in medical image classification. These advanced deep learning techniques have shown great promise in accurately diagnosing various medical conditions from imaging data. This study explores the application of CNN transformer-based models for the detection and classification of RA using thermal imaging—a passive, non-invasive technique that detects inflammation in the joints by capturing heat patterns. Thermal imaging is a passive, non-invasive method that uses the hand region's joint inflammation to identify RA. The study used 600 thermal hand images of healthy and RA patients to diagnose RA. The current study aim’s to implement a vision transformer (ViT) to identify RA from thermal hand images. The ViT classified the healthy and RA patients with an accuracy of 90%, with a precision, sensitivity and specificity of 0.83, 0.96, and 0.85 respectively. The results of this study underscore the potential of ViT models in the early prediction and diagnosis of RA, offering a non-invasive, efficient, and reliable method for identifying the disease in its early stages. This approach could significantly enhance clinical decision-making and improve patient outcomes by facilitating timely intervention.