Main comments:

In this manuscript from Karami and colleagues a newly tuned version of EC-Eath3 at high resolution (EC-Earth3-HR) is presented. The authors compare this new model version with observations, with the ensemble of EC-EARTH3 at standard resolution, and with the previous less tuned high-resolution model performed for HighResMIP. In addition, there is a special focus on understanding AMOC changes by analysing the deep mixed layer and deep-water formation. This is a very nice set of simulations, and its description and evaluation should certainly be published. The focus on the North Atlantic, the Arctic and the AMOC allows a deeper analysis for these regions and processes which is a good idea. However, before I endorse the publication of the manuscript there are three major points that should be improved.

First, since it is argued by the authors that the tuning is the difference with the previous high-resolution model version, I suggest that the authors explain the tuning better and reflect on it in the discussion section. For the tuning of the atmosphere, which sea surface temperature is used as boundary condition? Are the greenhouse gases set to 1850 values? What does “minimising the climate drift” mean? Which variables should not drift? How are the 5 parameters in Table 1 chosen? Are those known to be particularly important based on previous experiments with this model? How are those parameters influencing the radiative balance? Was the value of other parameters also tested?

Some more information is also needed for the coupled model tuning. For the choice of advection scheme, what does “overall better performance” (l.139) mean? Is it the results that are better? Were specific metrics used or did the authors look at maps of key variables to make a choice? Or was it the computational time or stability of the model that was important for this choice?

L.139-141: “The turbulent kinetic energy (TKE) mixing below the mixed layer was set to zero (nn_etau=0), as in EC-Earth3-SR (D.scher et al., 2022), which would otherwise lead to a significant reduction in AMOC”

Given that the AMOC changes between the time it is tunned (1850) and the time it can be compared to the RAPID array this is challenging. Was there a specific target for AMOC strength? The authors mention in the results section that the AMOC is larger in the model than in the observations. In retrospect should that value have been non zero?

The eddy diffusivity for tracers was increased a lot compared to EC-Earth3P-HR. What was the rationale behind that choice? It is rather counter-intuitive that this parameter is now the same in the standard resolution and in the high resolution because with a higher resolution one would expect that less eddy diffusion needs to be parameterised.

What about the viscosity? Which type of viscosity and which coefficient is used? How does it differ between the different model versions and resolution?

The second point to improve is the connection between the main conclusions and the analysis. The abstract is written in a precise way but in the discussion and conclusion section some claims are not supported by the analysis:

l.551-552: “the updated configuration shows reduced radiative imbalances and minimal deep-ocean drift.”

This is not shown. I suggest adding a figure to show it. There is a clear drift of the surface temperature in the model (Figure A1) so I expect that the deep ocean will also still be drifting.

The enhanced abilities or improved performance of the HR model compared to SR is not argued in the result section. Yet the following is written in the discussion and conclusion section:

l.559-561: “Time series of global and Arctic temperatures, AMOC strength, SST and sea ice area show better agreement with observations and reanalysis, indicating that higher resolution enhances the model’s ability to capture transient responses and low-frequency variability”

l.568-569: “The improved performance of the EC-Earth3-HR model in simulating Arctic–North Atlantic climate extends to key components such as Arctic sea ice, deep convection, and AMOC variability”

l.586-587: “This is while the horizontal and vertical distribution of water masses improves in EC-Earth3-HR.”

Based on Figure 6b it seems difficult to argue that HR performs better than LR for the Artic temperature. The improvements in AMOC trend and variability are also difficult to see Figure 9a. I suggest that the authors write more precisely what improves and what doesn’t in the HR model and define clear metrics to backup claims of improvements: rate of change, standard deviation, root-mean-square-error…

The third aspect to be improved is the data and code availability. This is important for other researchers to build on this work. The code used for the analysis and plot of this manuscript is not available. It is mentioned that the CDFTOOLS are used for the DWF analysis but this is not precise enough. Which script from the CDFTOOLS were used? With which options? Enough information should be provided for someone to be able to reproduce the DWF analysis. Especially since this analysis is a novelty from this manuscript. Concerning the data availability I think the following claim is misleading:

l.615-616: “Data from the EC-Earth3-HR historical and SSP2-4.5 simulations are available through any ESGF data node as part of CMIP6”

Isn’t this data from the HighResMIP version of the model (called EC-Earth3P-HR in this manuscript)? How and where is the data from the simulations analysed in this paper available?

Minor comments:

l.148: “the freshwater correction value was slightly modified to oas_mb_fluxcorr=1.07945”

Modified compared to what? The value is the same in the two models compared in Table 2. Also, a short explanation of what this parameter is would be useful. The name “oas_mb_fluxcorr” is not used in Doscher et al. (2022). Is this needed because IFS does not conserve water? Heat is also probably not conserved in IFS, is it also corrected for?

l.160-162: How long is this “long concatenated pre-spin-up run”?

Fig. 1a: Could you compare with longer reconstructions of TAS? For example, ERA5 is now available from 1940 while here only data from around 1980 is used.

l.243: “CAA” is not defined.

l.321: “PrI” is not defined

Fig. 9b: Is that the density change at the surface? It would be good to mention it in the caption.

l.412: It is not clear what “potential” means here. I would understand if “potential” was used for a claim about the real world but here it is about models.

l.464 and l.533: These correlation coefficients need more context. Are they computed on the yearly averaged data? Are the time series detrended? How? The trend is not linear. Are the correlations reflecting the co-variability at inter-annual time scale or a trend common to the time series?