This is a fairly comprehensive investigation of an important topic that is overall well presented, especially for a challenging topic to present like Arctic Ocean modeling. The conclusions are clear, but could be quantified a bit more (see major comment below); I recommend publishing this manuscript after revisions.

My largest comment is about how the results are quantified. The individual model biases are quantified, but the improvement or lack thereof between low and high resolution simulations is not, and in most cases is only qualitatively described. Is there a way to quantify how much a simulation improves for a given metric? The statements similar to ‘4 out of 5 models improved’ are technically quantitative, but I think going beyond this is worth the effort. A few specific examples:

• Figures 3-5 and 8 in particular should be better quantified. Why not plot Figure 3 as a bias the way Figure 7 is? Figure 7 explicitly plots the bias, and is an example where the improvement from low to high resolution is more obvious and quantifiable.
• Figure 11 and lines 278-295: It’s difficult to tell quantitatively which simulations agree best with the observations. What is the magnitude of the improvement (e.g., 100% error to 50% error)? Only a suggestion: can a basin-wide average warming be added in as a number on each sub-plot or as a separate table?
• Lines 402-403: CMCC-NEMO, quantitatively, which is better the low or high resolution case? Is there still a small improvement for higher resolution?
• The introduction is also not quantitative in terms of past studies. At some point it doesn’t matter if there is 100% error or 200% error because it’s wrong either way, but I was left wondering exactly how bad simulations are (apologies for being pessimistic).
• Overall, there is a lot of time spent discussing the individual models and not enough time spent discussing the improvements. It is clear that the models need to be discussed individually and the errors pointed out, but can the balance be shifted a bit more to discussing potential improvements? 

Additional comments

• Sea ice: How does increasing resolution effect the representation of sea ice? Better representation of leads or individual floes? Has this been addressed in other papers? Lines 663 – 669 about sea ice should be earlier in the paper, prior to the results section.
• Line 75-81: forcing is from 1958 to 2018, and this is repeated 5 times (so 300 years). By only analyzing the first 60 years, do you expect the Atlantic water to be stable / accurate? Is there one of the 5 pairs where the first and last cycles could be compared for AW properties?
• Why these metrics? Are these the metrics that can be easily tested? What other metrics would be good to have but can’t be compared with observations? Some introduction to this as a methods section or at the beginning of the results section would be appreciated. I also suggest explicitly stating that high-resolution metrics are not considered because the focus is on improvements from low to high resolution.
• Surface circulation has been mentioned a few times in the paper (e.g., lines 407-408), but it is not explicitly considered. Could it be included, maybe by comparing to Armitage et al. circulation products (even if this is a different year grouping than the PCH climatology)?
• Mixed layer depth comparison: Consider switching Figure S3 and Figure 12. Lines 325 – 327 suggest that the Figure 12 comparison to observations is not valid and certainly has fewer cautionary remarks than the Schmidtko comparison in Figure S3.
• Conclusions: To me it seems that conclusions #1,2, and 5 all boil down to gateway transports, and these metrics related to gateway transports are improved in high-resolution simulations. Conclusions #3-4 are about the ability to represent salinity accurately, and high-resolution does not always help with this. Is this an oversimplification of the findings? I might leap to saying that parameterizations of gateway transports should be the focus in terms of improving low-resolution models. Yes?

Minor comments

• Title: ‘higher resolution’ or ‘increased resolution’ seem less awkward to me.
• Line 89: ‘updation’?
• Line 183: This was confusing. Maybe clarify that intermediate and deep layers refer to the water beneath the 0°C AW lower boundary.
• Line 212: Primarily in the Eurasian Basin. Only improvement for the Amerasian Basin is NEMO, and ever so slightly HYCOM.
• Figures 4, 8: Why is 400 m depth chosen as the depth to investigate geographic patterns?
• Line 250: is there a way to reconstruct what the total mixing from all sources is for these models (explicit, applied parameterizations, and numerical)? How different would the total mixing be?
• Line 270-272: Reference to ‘model drift’ and ‘warming drift’ is confusing. Does this mean warming at depths below the AW core? Expansion of the AW core in depth? Please clarify.
• Figure 7: I think some version of the PHC3.0 salinity profile should be included in the main paper, and I would recommend including Figure S1 as is.
• Figure 12 and line 313-314: Are the observations also November to May averages?
• Figure 13: Is this computed from averaged fields, or individual fields and then averaged?
• Figure 14: Please clarify in the caption that positive values correspond to a deepening of the cold halocline depth (depth is positive downwards?).
• Figure 3, 16: For basin averages, is the average taken over the entire area shown in e.g., Figure 15, or is a certain section of this region chosen?
• Line 485: ‘warming of the Pacific Water’ by this you mean water in the Pacific Ocean and not Pacific Summer Water or Pacific Winter Water in the Arctic? (This manuscript doesn’t actually show that the Pacific Ocean water is warming, only that the heat transport is increasing, a reference on this could be useful too).
• Line 530: ‘improves noticeably’ should be ‘increases noticeably and is likely an improvement’? We do not know what the actual interannual variability is? So it is difficult to say if the lower resolution of higher resolution is more accurate? Is this comment made assuming that high resolution models are closer to the ‘truth’?
• Line 553-554: this statement about model spread should be stated earlier in the paper, as it is true for many of the variables considered here.
• Line 568-569: this would be good information to have in the introduction prior to reading through the results.
• Line 653 to 662: is some of the ‘improvement’ in interannual variability of gateway transports related to the improved simulation of eddies? Is there another cause for this change in interannual variability that should be speculated about?

Please see my comments in the attached pdf.