Tumors, which can be both hazardous and malignant, are a common health concern, and if left undiagnosed, they can significantly reduce life expectancy. Early tumor detection is a critical first step in developing an effective treatment plan. Brain tumors, in particular, pose a serious threat, as failure to detect them early can result in fatal outcomes. Various deep learning techniques have been proposed for brain tumor detection, with convolutional neural networks (CNNs) being one of the most widely used models for image classification tasks. This study explores the performance of the vision transformer (ViT) model, which we applied to identify different types of brain tumors. The study uses magnetic resonance imaging (MRI) scans as input data. To improve accuracy and precision, we implemented data augmentation techniques to expand the training set, while the vision transformer model was employed to extract key features from the images. The dataset used in this research combines three distinct sources: “figshare,” “SARTAJ,” and “Br35H,” containing 7,023 MRI images of human brains for model training. The images are categorized into four groups: pituitary tumors, gliomas, meningiomas, and non-tumor cases. The performance of the proposed model was evaluated based on accuracy, precision, and recall metrics. The vision transformer achieved an impressive accuracy of 98%, a precision score of 99%, and a recall rate of 97%.