Brain tumour detection and classification are crucial for accurate diagnosis and timely treatment, significantly impacting patient survival and quality of life. Early and precise identification of tumour type aids in selecting appropriate therapies and improving outcomes. This paper proposes a vision transformer (ViT) model for detecting and classifying brain tumours, leveraging a comprehensive dataset comprising multiple tumour types. The proposed model was assessed using performance indicators including recall, precision, F1-score, accuracy, and confusion matrix. It achieved an impressive accuracy of 90.39%, surpassing conventional approaches. By capturing localized and broad-scale features in medical images, the ViT model demonstrated high efficacy, particularly for non-tumour and pituitary tumour classes, while facing challenges in differentiating glioma and meningioma due to overlapping features. The study also highlights the ViT model’s strengths, including its ability to identify long-range dependencies and its resilience against noise. These findings emphasize the promising role of ViTs in medical imaging for advancing diagnostic methodologies and improving patient outcomes.