The authors describe an approach for analysis of inter-dependence of CMIP climate model simulations, their weighting and sub-selection for different purposes based on desired range of spread, performance and dependency.

The manuscript is well written and fits the scope of GMD. I find the developed methodology innovative and useful.

One of the important findings is that the model performance is rather a “model” characteristic, whereas the spread is more diverse for individual members of an ensemble of the same model. Further, the reduction of the spread of ECS after the family-democracy is taken into account is also a very important conclusion.

Please find below comments that should be addressed before the paper is accepted for publication:

line 234 – 237: I suggest explaining better that the “multi-model ensembles” correspond to “families”, e.g. replacing the word “ensembles” with “families”.

line 260-265: the results of the sensitivity testing are shown somewhere? it should be stated explicitly (e.g. “see below”)

line 270 – 274: I suggest shortly mentioning that the benefit of longer time period is not visible for all models, denoting the contradictory result of EC-EARTH3 models, for which there is still the overlap even for the longer period.

line 294: I suggest adding a note that the concept of fingerprints will be explained further below.

line 546 – 547: why the evolution of SAT over Europe should be representative of the GCM’s ability to simulate correctly the response to aerosol forcing? There are also other factors to be taken into account, so why specifically only the aerosol emissions are mentioned here?

line 608: I recommend to explain a bit the term “pool” – it can be the whole multi-model ensemble or somehow pre-selected subset. The term pops-up suddenly and makes the reader a bit confused.

lines 612-620: the notation "s_i" should be explained properly, that it denotes individual simulations.

line 875: please add a reference to the proof of the statement “intermember distances within both CMIP ensembles did satisfy metric criteria”. (is it shown somewhere or not shown?)

Part II – a comment on ternary plots and recommended subsets: The ternary plots are definitely useful for the analysis of different selection criteria. An issue, that is not commented on, is that some of the subsets “reside” a large part of the triangle, whereas some other subset have only a small fraction of the triangle. In some cases, the subset minimizes the cost function for only very narrow intervals of the coefficient values. I suggest that it should be discussed, that in the case of the subsets that correspond to a very small fraction, there might be other subsets that have cost function values close to minimum and would maybe satisfy the criteria for a wider interval of the coefficients? Could there be some additional selection criteria that the recommended subsets should minimize the cost function for a larger fraction of the ternary plot? It would make the selection more robust. In some cases, the recommended subsets are represented by only several “points” in the ternary plot (e.g. Fig 8). I believe that it is desirable to recommend subsets that would be useful for as wide range of applications as possible, to make projections used for similar applications physically consistent.

Part II + Discussion and conclusion: Regarding the recommended subsets derived from CMIP5, the ClimSIPS method suggests similar subsets as used in Euro-CORDEX for driving regional climate model simulations over Europe. The authors claim, that this agreement implies, that their method is suitable for choosing subsets for driving RCMs. This implication is questionable, as it is not clear, what exactly was the basis for the choice of Euro-CORDEX driving GCMs from CMIP5. I do not doubt that proposed method is suitable for choosing appropriate subsets from CMIP6, I just do not agree with the comparison to CMIP5 subsets implying the suitability of ClimSIPS. Please, consider modifying the statements appropriately. The argumentation should be based on the nature of ClimSIPS, which is well described in the paper.

Language, copy-edits:

line 85 – Sentence beginning „Initial versions ...“ – the verb is missing in the sentence.

line 149 – „The study, an extension of the work ...“ – the “of” is missing

line 508 – “in” is missing in “For use in cases...”

This manuscript introduces a method (ClimSIPS) which select subsets of CMIP models based on model independence, model performance and spread. In the second part the authors describe a case study for European summer and winter.

The manuscript fits the scope of GMD and is very helpful by dealing with the large ensemble of CMIP5 and especially CMIP6 models. Additionally, the change of the ECS distribution when considering only one model family member is very interesting. Nevertheless, the manuscript has reached an extreme length and is written very detailed. I am wondering if there are places where the text could be shortened.

Some small comments:

Line 53: Capital letter in the beginning of the sentence: “Modeling centers…”

Line 86/87:  A verb is missing in the first part of the sentence.

Figure 1: The quality of the figure is quite bad and difficult to read.

Line 568: Is there a special reason for choosing this reanalyse dataset?

Line 945 and 947: Is this grade of precision of the numbers really needed here?