Earth system models (ESMs) are essential tools for understanding the climate system and projecting future changes, but the standard coarse resolutions often fail to realistically represent regional processes and topography, particularly in the Arctic. To address this, regionally refined meshes (RRMs) have been developed within ESMs to provide high-resolution simulations in target areas, improving the accuracy of regional climate simulations while maintaining computational efficiency. This study provides an overview of the Arctic regionally refined model (RRM) configuration of the fully coupled Earth System Model version 2.1 (E3SMv2.1), hereafter referred to as E3SMv2.1-Arctic, based on historical simulations (1950-2014). We make comprehensive evaluations focusing on the atmospheric component and the interactions among the atmosphere, land, ocean, and cryosphere in the Arctic through comparisons with the low-resolution (LR) configuration of E3SMv2.1, reanalysis products, and observational datasets. In the Arctic region, the RRM (atmosphere model resolution of ~0.25°) generally reduces biases present in the LR model (standard resolution of 1.0°), showing improved simulations of large-scale fields, such as precipitation, low-level atmospheric circulation, clouds, atmospheric river frequency, poleward intrusion and generation of strong cyclones, and land water storage change due to better-resolved topographic effects, surface thermal contrasts, land-river connection, and other key processes. The RRM also demonstrates a seasonally dependent bias in the surface air temperature, with a reduced cold bias in summer but an enhanced warm bias in winter compared to LR. Sea ice analysis reveals that the RRM underestimates winter sea ice area and volume, consistent with the strong winter warm bias, while LR overestimates both. Seasonal sea ice in RRM grows faster and melts more rapidly than in LR, primarily due to the warmer sea surface temperatures and the resulting stronger basal melting. Radiative feedback analysis indicates that both RRM and LR exhibit similar climate feedback strengths, with the main difference being a more positive surface albedo feedback in RRM, contributing to a stronger surface warming. Finally, the RRM simulates more numerous strong cyclones in the central Arctic compared to the LR, with a positive bias compared to reanalysis. These findings highlight the importance of high-resolution modeling in improving our understanding of Arctic Earth system changes and their global impacts, although some persistent biases appear to be independent of model resolutions at 10 - 100 km scales.