Brain tumors are the biggest medical challenge that requires more diagnosis and treatment planning to identify the disease progression for that segmentation used to determine accurately growing cells. The proposed architecture is based on a combination of the transformer model, and the decoder part is adapted from the UNET architecture followed by the attention gate for selectively focusing on features and improving prediction accuracy. The BraTS2020 dataset, which is used for the experimental analysis, has four different modalities as MRI images for a variety of glioblastoma and low-grade glioma cases. After applying the Attention-based UNETR model on the BraTS2020 dataset for segmentation, the evaluation parameters are considered as the dice score coefficient which gets results for enhancing tumor, core tumor, and whole tumor are 0.7727, 0.8619, and 0.7629, respectively. The result demonstrates the model’s robustness, making it a promising tumor segmentation method.