Segmenting a tumor accurately from a multimodal image in medical imaging is critical for the right diagnosis and therapy. This research comprises efficient and precise tumor segmentation using unique 3D U-Net architecture on multimodal imaging data, mainly on MRI modalities. The approach that has been proposed comprises aspects like attention, residual, and a multimodal architecture for the goal of segmentation. The BraTS 2021 dataset, which comes with multimodal MRI scans of patients together with voxel-level tumor annotations, was used for evaluation of the built model. Several common metrics were employed for analyzing the suggested model, including Dice Similarity Coefficient (DSC), Intersection over Union (IoU), sensitivity, and specificity. The DSC, IoU, sensitivity, and specificity of the suggested 3D U-Net were 0.92, 0.86, 0.94, and 0.91, respectively. These results were better than those of baseline approaches like Attention U-Net and Residual U-Net. Results showed improvement in accurately determining tumor boundaries and coping with intricate tumor forms. Thus, this work established the potential of the suggested 3D U-Net architecture for tumor segmentation in multimodal imaging, which can be employed in clinical applications like radiation planning and surgical navigation. The approach might be extended to other forms of cancer, and lighter models could also be produced for further studies.