Recent advances in data-driven control systems have highlighted the adaptability of Gaussian Process Regression (GPR) in managing complex, nonlinear dynamics. Unlike traditional approaches that sequentially conduct system identification and model-based control design, this paper introduces the Variance-Informed Model Reference Gaussian Process Regression (MR-GPR). This method integrates GPR directly into a model reference framework using input/state data to infer inverse dynamics, enhancing control strategies through the utilization of predictive variance. Applied to discrete-time nonlinear systems, the variance-informed MR-GPR approach effectively adapts to changing conditions by leveraging predictive variance and real-time feedback, thereby enhancing control system robustness without the need for explicit system identification.