The medium temperature conveyed by the high-temperature molten salt pump for the reactor reaches up to 650°C. However, the core components of the upper driving end, including magnetic levitation bearings and the motor, need to operate within an appropriate temperature range. Therefore, controlling heat transfer and effectively suppressing the temperature rise of the driving end are key technologies in the research and development of molten salt pumps. Based on the Response Surface Methodology (RSM), this paper identifies the thickness of the outer cylinder, the thickness of the outer ring frame of the thermal insulation screen, the thickness of the inner ring frame of the thermal insulation screen, and the shaft diameter as key parameters affecting the temperature rise in the upper part of the molten salt pump. These parameters were ranked by significance, and an approximate model was established between the key parameters and the average temperature of the magnetic levitation bearing. Multi-objective optimization of the high-temperature molten salt pump was conducted using Non-dominated Sorting Genetic Algorithm II (NSGA-II), yielding a set of optimal schemes that meet the design requirements. The accuracy and reliability of the optimization model were verified through simulation calculations. Compared with the original scheme, the average temperature of the under magnetic levitation bearing in the optimized high-temperature molten salt pump has decreased by 10.3°C. Meanwhile, this optimization scheme has successfully ensured the comprehensive bending stiffness of the equipment, with both radial and axial stiffness indicators remaining within the designed threshold range, thereby further improving its safety and service life. The research results of this paper can provide valuable references for the lightweight design and large-scale development of high-temperature molten salt pumps.