Arctic Ocean simulations in 19 global ocean-sea-ice models participating in the Ocean Model Intercomparison Project (OMIP) of the Coupled Model Intercomparison Project Phase 6 (CMIP6) are evaluated in this paper. Our findings show no significant improvements in Arctic Ocean simulations from the previous Coordinated Ocean-ice Reference Experiments phase II (CORE-II) to the current OMIP. Large model biases and inter-model spread exist in the simulated mean state of the halocline and Atlantic Water layer in the OMIP models. Most of the OMIP models suffer from a too thick and deep Atlantic Water layer, a too deep halocline base, and large fresh biases in the halocline. The OMIP models qualitatively agree on the variability and change of the Arctic Ocean freshwater content; sea surface height; stratification; and volume, heat, and freshwater transports through the Arctic Ocean gateways. They can reproduce the changes in the gateway transports observed in the early 21st century, with the exception of the Bering Strait. We also found that the OMIP models employing the NEMO ocean model simulate relatively larger volume and heat transports through the Barents Sea Opening. Overall, the performance of the Arctic Ocean simulations is similar between the CORE2-forced OMIP-1 and JRA55-do-forced OMIP-2 experiments.