CAMPaS is a translational AI framework that enables unified glioma classification by jointly predicting histological grade and subtypes, and key molecular markers. It incorporates two specially designed modules, a bi-directional attention module and a noise-robust learning module, to address challenges arising from dataset heterogeneity and label uncertainty in real-world clinical data. Technically, CAMPaS is implemented as a multi-task, multi-instance learning framework built upon a hierarchical vision transformer backbone.