Soil heavy metal contamination is a major environmental issue threatening food security and human health, and mathematical models are key tools for quantitatively predicting the migration and transformation of heavy metals in the soil-plant system. However, model development in this field currently exhibits a significant "asymmetric pattern." This paper systematically reviews and analyzes the state of modeling research on heavy metal transfer from soil to plants, aiming to reveal the specific manifestations, underlying causes, and future directions of this asymmetry. The study finds that: 1) On the soil side, mechanistic models based on the coupled framework of "hydrodynamics–solute transport–geochemistry" (e.g., HYDRUS, PHREEQC) are already highly advanced. These models can precisely simulate the physical transport of heavy metals and their speciation transformations based on surface complexation theories (e.g., CD-MUSIC, NICA-Donnan), indicating that soil modeling has entered a stage of mechanistic and fine-scale representation. 2) In stark contrast, model development on the plant side lags significantly. Current approaches still largely rely on empirical parameters such as the bioconcentration factor (BCF) and transfer factor (TF), lacking mechanistic descriptions of critical physiological processes including membrane transport, long-distance translocation, and tissue compartmentalization—creating a pronounced "mechanistic gap." 3) At the system coupling level, a "disconnection" exists: most current models adopt a unidirectional, soil-to-plant linkage, neglecting dynamic feedback from root activities (e.g., root exudates, pH regulation) on the soil microenvironment, and thus failing to establish a closed-loop interactive system. To advance modeling toward greater integration and systematization, this paper proposes future research priorities: efforts should focus on resolving the coupling challenges at the "root-soil interface," developing dynamic heavy metal accumulation models based on functional-structural plant models, and establishing new paradigms for multi-source data fusion and model-data assimilation. These breakthroughs are essential for bridging the mechanistic gap and achieving bidirectional coupling and accurate prediction in soil-plant systems.