The Arctic Ocean is projected to experience not only amplified climate change but also amplified ocean acidification. Modeling future acidification depends on our ability to simulate baseline conditions and changes over the industrial era. Such centennial-scale changes require a global model to account for exchange between the Arctic and surrounding regions. Yet the coarse resolution of typical global models may poorly resolve that exchange as well as critical features of Arctic Ocean circulation. Here we assess how simulations of Arctic Ocean storage of anthropogenic carbon (Cant), the main driver of open-ocean acidification, differ when moving from coarse to eddy-admitting resolution in a global ocean-circulation–biogeochemistry model (Nucleus for European Modeling of the Ocean, NEMO; Pelagic Interactions Scheme for Carbon and Ecosystem Studies, PISCES). The Arctic's regional storage of Cant is enhanced as model resolution increases. While the coarse-resolution model configuration ORCA2 (2∘) stores 2.0 Pg C in the Arctic Ocean between 1765 and 2005, the eddy-admitting versions ORCA05 and ORCA025 (1∕2∘ and 1∕4∘) store 2.4 and 2.6 Pg C. The difference in inventory between model resolutions that is accounted for is only from their divergence after 1958, when ORCA2 and ORCA025 were initialized with output from the intermediate-resolution configuration (ORCA05). The difference would have been larger had all model resolutions been initialized in 1765 as was ORCA05. The ORCA025 Arctic Cant storage estimate of 2.6 Pg C should be considered a lower limit because that model generally underestimates observed CFC-12 concentrations. It reinforces the lower limit from a previous data-based approach (2.5 to 3.3 Pg C). Independent of model resolution, there was roughly 3 times as much Cant that entered the Arctic Ocean through lateral transport than via the flux of CO2 across the air–sea interface. Wider comparison to nine earth system models that participated in the Coupled Model Intercomparison Project Phase 5 (CMIP5) reveals much larger diversity of stored Cant and lateral transport. Only the CMIP5 models with higher lateral transport obtain Cant inventories that are close to the data-based estimates. Increasing resolution also enhances acidification, e.g., with greater shoaling of the Arctic's average depth of the aragonite saturation horizon during 1960–2012, from 50 m in ORCA2 to 210 m in ORCA025. Even higher model resolution would likely further improve such estimates, but its prohibitive costs also call for other more practical avenues for improvement, e.g., through model nesting, addition of coastal processes, and refinement of subgrid-scale parameterizations.