Numerous historical documents have been composed using the Modi script, and their preservation and digitization are crucial for enhancing public understanding of the history recorded in Modi. However, the digitization of Modi script documents has made limited progress. In this study, we introduce a vision transformer model for recognizing handwritten Modi script characters, aiming to surpass existing methods and improve accuracy. Our approach utilizes an advanced vision transformer architecture that consumes preprocessed data through grayscale conversion, random perspective transformation, and random affine transformations. The model consists of an embedding layer, a stack of transformer encoder blocks, and a classifier head. Training involves minimizing cross-entropy loss with the Adam optimizer, using a designated portion of the dataset for training and a separate test set for evaluation. The experimental results show that our model significantly outperforms state-of-the-art methods reported in the literature for handwritten Modi numeral recognition, achieving an accuracy of 97.35%, establishing a new benchmark with exceptional precision and accuracy. This study is pioneering in successfully applying vision transformer technology to handwritten Modi script recognition.