Brain tumors are among the most lethal neurological disorders, making early and correct diagnosis crucial to successful treatment. The improved soft-tissue contrast of Magnetic Resonance Imaging (MRI) makes it a popular tool for detecting brain tumors. MRI scan interpretation by hand, however, takes a lot of time and is prone to human mistakes. A hybrid deep learning model for autonomous brain tumor classification is suggested in this study that combines Transformers and Convolutional Neural Networks (CNNs) in order to overcome these limitations. While the Transformers use self-attention techniques to record global contextual relationships, the CNN component, which is based on a pretrained MobileNet, extracts local spatial data from MRI images. A publicly accessible “Brain Tumor MRI” dataset with four classes - glioma, meningioma, pituitary tumor, and no tumor is used to train and assess the model. After preprocessing the data, stratified sampling and the class weights balance technique are used to ensure balanced training. With an overall accuracy of 99.24 %, the model performed effectively in categorizing previously unseen brain tumor images. Additionally, it obtained a 99.30 % weighted F1-score, 99.30 % precision, and 99.54 % recall. These findings demonstrate how well the CNNTransformer hybrid model can be used to diagnose brain tumors using computer aids, with encouraging classifications for clinical decision support.