This study developed and validated a non-invasive multi-metabolite magnetic resonance spectroscopy framework for preoperative molecular subtyping of adult diffuse gliomas. Using a cross-center, cross-vendor cohort of 268 patients, 48 features derived from 18 metabolites were systematically evaluated to identify discriminative metabolic signatures for predicting IDH mutation and 1p/19q codeletion status. The integrated metabolic model showed robust intra-center and cross-center performance for IDH prediction, with AUCs of 0.906 and 0.857, respectively, and for 1p/19q prediction, with AUCs of 0.858 and 0.787, respectively. These results suggest that synergistic metabolic profiling support molecular assessment in patients who may not be suitable for invasive biopsy.