<p>The Korean Integrated Model (KIM) is a global numerical weather prediction model that uses the spectral element method on a (quasi-) uniform-resolution cubed-sphere grid. However, to investigate high-resolution prediction performance in the domain of interest, global high-resolution experiments must be considered. These experiments incur a significant computational burden owing to the increased number of grid points and the need for temporal integration at finer time steps. To address this issue, we implemented a stretched variable-resolution grid in the KIM, based on the Schmidt transformation. This approach allows for high-resolution effects in the region of interest while using a coarser grid elsewhere. With the stretched global grid, grid sizes are seamlessly scaled between high and low resolutions.</p><p>In this study, we evaluated the forecasting accuracy and computational efficiency of the variable-resolution KIM and compared these metrics with those of the corresponding uniform-resolution KIM. In the higher-resolution target region, the forecast skill of the variable-resolution grid was similar to that of the 8-km high-resolution uniform grid and even outperformed the 25-km reference-resolution uniform grid in the KIM. Furthermore, the variable-resolution configuration offers a significant improvement in computational efficiency, reducing the total 3-day forecast run time by approximately 88% compared with that of the high-resolution configuration with a uniform grid.</p>