This retrospective multicenter study explored the potential of CT-based radiomics to improve the preoperative risk assessment of gastrointestinal stromal tumors (GISTs), which are characterized by variable malignant potential, by capturing tumor heterogeneity beyond conventional imaging features. In 123 patients, tumors were segmented on contrast-enhanced CT scans and a large set of radiomics features were extracted and refined. Among the tested algorithms, a support vector machine model achieved the best performance. The model demonstrated strong predictive accuracy in both the testing and external validation cohorts, with good agreement between predicted and observed outcomes, and showed clear clinical utility in differentiating low- from high-risk tumors. Overall, the CT-based radiomics approach may provide a reliable tool for GIST risk stratification and support more personalized treatment strategies by identifying patients requiring additional therapy and those eligible for surgery alone.