General comments:

The paper provides a useful overview of the coupled AWI-CM3 model, providing a near direct comparison between two model resolutions, as well as evaluation against observations/re-analysis and CMIP6. Particular attention is given to polar climate, where increased resolution has a larger impact on mesoscale processes unresolved by coarser resolution. The comparison is between a pair of simulations from 1950 - 2100, with most of the emphasis on evaluation of biases relative to present-day observations or reanalysis, rather than on future projections. Overall this is a well-written and clearly understandable manuscript that provides a very useful benchmark for the AWI-CM3 model and the impact of increased resolution.

Specific comments:

Each model is run for 250 years from initialization in repeat 1950 conditions, with the first 100 years considered spinup and subsequent 150 years as control. It would be useful to have some mention of the model’s level of equilibrium during this control period, and/or see metrics. Figure 3 shows some of this; the global SST and 2m air temperatures seem to be in equilibrium in the control, but maybe some of the sea ice metrics have a trend during the control? Is the ocean steadily gaining or losing heat during the control? In Figure 7d the AMOC time series appear to be declining from 1950, was this trend also in the control?

This paper investigates the impact of model resolution in coupled simulations using the AWI-CM3 model, which is one of the CMIP6-participating models. The model was run in two configurations: a lower-resolution setup (LR; atmosphere: ~102 to 120 km with 91 vertical levels; ocean: CORE2, 20 to 100 km) and a higher, medium-resolution setup (MR; atmosphere: ~32 to 38 km with 137 levels; ocean: DART, 4.5 to 25 km). The higher resolution permits ocean eddy-resolving capability. The paper covers a variety of evaluations to examine performance changes due to the increased resolution. The authors find that increased resolution leads to improved model performance overall. I believe the paper is well written and the scope fits the journal. However, before publication, readers would benefit from a revised manuscript, including the following.

Section 2: What is the simulation period? I assume that is identical to the CMIP6 Historical experiment protocol, but it may be worth noting it clearly in the text.

Table 1: I suggest adding the equivalent resolution information in km for the atmospheric and ocean grids, which will help improving the clarity of the configuration.

Fig. 1: I suggest use “LR” and “MR” or “lower” and “higher” for the 2^nd and 3^rd column titles, as the readers may not familiar with “CORE2” and “DART” terminologies thus have to revisit the text where they were defined.

Fig. 2: I suggest adding a color bar. Also, readers would benefit from knowing exact variables used in the figure. Some abbreviated variable names on the y-axis may not be very straightforward especially for those who are not in the field (e.g., mlotst, thetao, so, etc), although some were described in the text (line 155-170). I suggest providing full names of the variables (recommended) in a comprehensive way (like a table in the supplementary) and point to where they could be found from the figure caption.

Line 151-153: Regarding the overall score, some arbitrary is unavoidable depending the selection of variables and regions, and there has been long debates about defining the “overall score” in the community. I recommend discussing about such limitation of the overall score.  

Fig 6: I find a mismatch between the figure and the interpretating text, and it puzzles me. It looks like the color bar to be the opposite of the interpretation of the result in the section 3.3. For example, lines 254-255 describes about the observed sea ice loss in the central Arctic (Fig. 6g) where in the figure, most of the Arctic sea is colored in red, in which according to the color bar, it indicates positive SIC trend, thus increasing sea ice. I encourage the authors double check the figure to make sure what it shows.  

Fig 11. I recommend using the same configuration name (LR, MR) throughout the paper for consistency. I also wonder the lines from LR, MR, and OBS could be superimposed to same space and make comparison easier.