In this paper, Ed Blockley and his co-authors provide a description of the sea-ice model GSI9 and its coupling within the Coupled Model GC5.

The paper is clearly structured, provides a helpful summary of the functioning of the model, describes in detail the technical aspect of the coupling of GSI9 to the ocean and to the atmosphere, and is very well written. I therefore generally recommend its publication, but I think that the following comments should be addressed in a revised version. They are in particular geared towards increasing the usefulness of this paper for a general sea-ice modelling audience who would like to draw insights from such paper for their own work.

Overall, I felt that the paper provides too little guidance to the reader regarding the motivation of the shift to SI3 and to the new coupling scheme. A description of the advantages and disavantages of SI3 vs CICE (if possible) and of flux coupling vs. standard coupling would be helpful. The reader could then infer for themselves whether such shift is considered useful for scientific / numerical / strategic reasons.

I also would have liked to see a broader discussion of the performance of the new model setup. How computationally expensive are these runs relative to the ones with the previous model version (or: Which percentage of the ocean computations happen within the sea-ice module in this version and in the previous version)? How well does the sea-ice model scale in this setup with the number of CPUs compared to the previous model version, unless this is documented elsewhere?

For the coupling, more information would be helpful regarding the standard coupling frequency and standard time step for the atmosphere and the ocean. I was surprised to read that coupling occurs at 2-3 ocean and atmosphere timesteps, as this seems to imply a similar time step in the atmosphere and the ocean. Is this indeed the case? Most models that I am aware of use a much longer time step in the ocean than in the atmosphere. Are there any drawbacks for a lower coupling frequency in the flux-coupling approach compared to the standard approach, given the somewhat unphysical development of internal sea-ice temperature for a fixed surface temperature between the coupling intervals? Or is the flux coupling of advantage, as the surface temperature and the atmosphere interact physically more realistically, and this is considered more important in a coupled setup?

Regarding model evaluation, the current analysis does not allow too much insight regarding the source of the potential model improvement, and in particular little insight into potential improvements related to the sea-ice component. For the Southern Ocean, improvements in the ocean component rather than in the sea-ice component are mentioned as the primary reason for the improved sea-ice simulation. It remains unclear whether improvements e.g. in the atmosphere component are the primary reason for the improved sea-ice thickness distribution in the Arctic. Is there any indication that the sea-ice component itself has contributed to these improvements in either hemisphere? I know this is hard to show, but maybe an analysis of more sea-ice inherent properties would be more helpful (e.g., melt-pond fraction, small-scale variability, lead fraction, or the like).

Finally, I was not sure why HadISST was used for the sea-ice concentration comparison rather than one of the standard direct satellite products. Using HadiSST 2.0 can be misleading, as they use a very generous ocean grid with many islands etc being water. "This results in much larger sea ice extents in HadISST.2 for all calendar months, unless the same mask is applied. We recommend that the same grid and mask are used when comparing any sea ice concentration data set." (quote from HadISST 2.2.1.0 website)

Minor comments:

l.69: What does "largely similar" mean? Shouldn't it be either "similar" or "largely equal"? ;)

l.76: Are the uppermost ocean grid cells completely turned into ice as the ice approaches very large thickness? Which coordinate system is used in the ocean model and how is the ice incorporated into it?

l.89: Is this a different convection scheme compared to previous model versions? If so, what was used before? Is third-order necessary and/or helpful?

l.96: Not sure what "this" refers to

l.128: Would be helpful to briefly indicate how the heat-flux calculation by Maykut and McPhee (1995) works

l.136: What is "the beta coefficient"?

l.147: Is salinity also used to calculate energy content / heat capacity in the thermodynamics?

l.226: How relevant is passing these velocities to the atmosphere, as they should be relatively small compared to the wind speed. Is the numerical overhead of passing them over negligible? Or would setting them to 0 in the atmosphere suffice? (Maybe nothing to be examined for this paper, but maybe this is known)

l.302ff: I would recommend to leave this out, or to move it to somewhere else. It's neither a conclusion, nor really helpful at this stage, I find. Maybe it'd be better to integrate the individual future plans into the relevant sections of the paper, with a brief motivation.

Typos / Grammar

l.47: Drop "is"

l.48: "are discussed in the  GC5 paper"

l.80: Drop comma

l.90:  'or "to" open water' is somewhat easier to read, I find

The authors present a detailed description and limited evaluation of GSI9, an upgrade to their sea ice coupling scheme for UKESM. While their previous GC3.1/CMIP6 setup employed CICE in both the ocean model NEMO and the atmosphere & land model UM-JULES, the new GC5 prototype for CMIP7 use replaces CICE on the ocean side with the NEMO native sea ice model SI3. The conductive coupling approach, which places the atmosphere-sea ice coupling interface between the top and second layer of sea ice rather than between the top layer and the atmosphere, remains unchanged.

The technical description of the work is clear and concise, and the idiosyncrasies of the conductive coupling approach are well-documented in the cited literature. However, I would like to see a paragraph on the motivation for using SI3 for NEMO4 rather than continuing to use CICE. Additionally, the manuscript would benefit from detailing why the advantages of using SI3 outweigh any potential advantages of maintaining a consistent sea ice physics formulation with CICE on both sides of the coupler.

Although the authors state that a detailed analysis of the resulting sea ice climate is not within the scope of this paper, I encourage them to broaden the scope slightly to include a basic characterization of the sensitivity of the old and new sea ice schemes different climate states. One approach could be to run a short 1850 control simulation followed by a 1% CO2 increase per year, allowing for the computation of transient climate response (TCR) and sea ice response. Another option may be to approximate the CMIP6 HighResMIP protocol (as forcing implementation permits) and show the Arctic Amplification Indices and sea ice response. Other approaches can yield similar information. I consider this relevant as UKESM GC3.1 was an outlier with the highest climate sensitivity CMIP6 dataset.

Finally, it would be beneficial to include a discussion, or provide a reference if analyzed elsewhere, on the phase error in the Arctic summer sea ice area minimum, which occurs in September in observations but shifts to August in both UKESM versions., See Figure 2. The improvements in the Southern Hemisphere, on the other hand, are very encouraging, even if much of it may be the result of ocean modelling improvements.

Overall, I recommend the paper receive a minor revision before acceptance.

Minor Comments:

• L525: Why is second-order accuracy used for only two out of the nine radiative fluxes?
• Appendix A, page 24: Snow volume is halved after ridging (as noted in the text), yet the melt pond fraction remains the same (rn_fpndrdg=1). While this may be good for tuning, it seems incorrect from a practical perspective. If the snow falls off, so should the liquid water.
• Appendix A, page 24: The text mentions increasing the number of layers from 2 to 4, but here it says the layers are reduced from 4 to 2. Please verify this information.