High-resolution climate projection datasets for South America require careful model selection, since regional climate depends on interacting atmospheric systems across multiple scales. Consequently, evaluations based only on statistical metrics may fail to determine whether models realistically reproduce the key phenomena organizing climate over the continent. As the first step in building the HiClim-SA dataset, this study proposes a novel framework for selecting models from the Coupled Model Intercomparison Project Phase 6 (CMIP6). Evaluation metrics are derived from objective criteria defined for major atmospheric phenomena over South America and integrated through a multi-criteria ranking applied to 19 CMIP6 Earth System Models. The results revealed substantial differences in model skill across the evaluated phenomena, indicating that no single model performs best for all aspects of South American climate. Instead, only a limited subset of models emerges as consistently more robust across precipitation, convection, and atmospheric circulation, whereas the lowest-ranked models exhibit broader and more persistent biases over tropical and central South America. Comparisons with baseline rankings showed partial agreement for the best- and worst-performing models, but also important discrepancies for several intermediate models. These findings demonstrate the added value of the proposed framework by showing that process-oriented model evaluation can identify robust model results beyond what is captured by conventional climatological metrics alone. The framework therefore provides a model selection based on the analysis of multiple atmospheric phenomena and establishes the methodological foundation for the subsequent bias correction, downscaling, and future projection stages of the HiClim-SA dataset.