<p>Three-dimensional mineral prospectivity modeling (3D MPM) is an emerging tool for targeting deep-seated concealed mineralization. Recently, three-dimensional convolutional neural network (3D CNN) has gained increasing application in 3D MPM due to their ability to integrate multi-source 3D predictive maps. However, existing 3D CNN-based models typically require extensive preprocessing of 3D geological models before they can be integrated with continuous predictive maps. This preprocessing often relies on artificially selected spatial analysis methods to quantify the influence of geological structures on mineralization. While useful, this approach may not be able to fully extract the ore-controlling features from the 3D geological models and may result in a potential risk of acquiring ineffective information by the deep learning model. To address these issues, this paper proposes a hybrid 3D GCN-CNN model that combines the 3D graph convolutional network (3D GCN) module for modeling distance relationships among geological structures with the 3D CNN module for extracting geometric morphology and spatial distribution patterns of geological bodies. This hybrid framework directly extracts information from 3D geological models without spatial analysis. The proposed methodology is applied to a case study in the eastern Chating area in Anhui Province, China. By employing the 3D GCN–CNN hybrid model, this paper demonstrates its superior performance compared to the 3D CNN model. Drilling validation confirmed the effectiveness of the 3D GCN-CNN model, suggesting that the proposed framework offers an efficient method for deep mineral exploration.</p>