I appreciate the authors attempts to revise the manuscript according to the reviewers comments. However I feel that while it is useful to have all 46 CMIP5 models evaluated and referenced someplace, and have both the Arctic and Antarctic evaluated, the current manuscript is not up to the standard for a manuscript in The Cryosphere.
Part of the problem stems from the methodology. First off, only looking at extent and/or volume is not all that useful. The spatial maps shown are helpful for the ice concentration, but what should also be shown are the spatial maps of the CMIP5 thickness fields. A recent paper by Stroeve et al. (2014) published in The Cryosphere already showed that while the Arctic ice volume in CMIP5 are comparable to that in PIOMAS, the spatial pattern of where the models have thick vs thin ice is not well represented in the CMIP5 models. I also feel that you need to show in this paper that the standard deviation between different ensemble members from the same model are comparable, not just in terms of total extent and volume, but also the spatial patterns. The Turner et al. (2013) paper only looked at the Antarctic and they didn't use as many models. Since one of your points for why this paper is important is that all 46 models are evaluated, then I feel that more robust analysis of the variability for all those models is required. Also, some discussion as to the 1979-2005 time-period is warranted. While it is the period of observations (and therefore, a useful comparison), it is additionally important to discuss the results in the context of not expecting the models to necessarily be in phase with the natural variability seen in the observations. This is likely very important in your discussions of Antarctic sea ice.
It is not correct to state that sea ice thickness data are primarily ship based observations. Do you mean submarine data? There are also thickness data from aircraft and satellite. Stroeve et al. (2014) compared the thickness data in the Arctic with that of PIOMAS and reasoned that PIOMAS provided useful estimates of ice thickness/volume to compare with the CMIP5 models. Can you not make a similar assessment of the accuracy of GIOMAS for the Antarctic? Some basis for using GIOMAS is needed here.
There should be some statistical assessment as to whether the CMIP5 MME trends in SIE and SIV are statistically different from those observed and modeled by PIOMAS and GIOMAS.
The Tables could be made more useful if the models that had statistically different trends were highlighted somehow (i.e. via shading)
In general, I do not find that much new information on Arctic and Antarctic sea ice representation in CMIP5 models is given beyond previous studies. Figure 11 may be the most interesting part of the paper and should be expanded upon.

Lines 63-66 need to be rewritten, it is incorrect to say the CMIP5 could reproduce the observed Arctic ice-free time, as the observations do not show ice-free conditions.
At the end of the Intro I still think you need to give more justification for the paper than just showing more models. What new knowledge is gained in this assessment? This should be mentioned here.
Lines 83-86, it would seem appropriate here to reference your analysis of the 15 GISS ensembles to support your assertion that the ensemble spread between models is small. It is my understanding though in the large CESM ensemble that the spread in those ensembles is on the same order as between all CMIP5 models, so you may also want to comment on that (Work by A. Jahn). Thus, while you see little variability between the 15 GISS ensembles, the large (100 ensemble) CESM suggests otherwise.
lines 96, it seems appropriate here to reference studies that have "validated" PIOMAS, and mention that no such similar analysis is available for GIOMAS.

There remain instances that the language could be improved by a native english speaker, such as changing statements such as "standard deviation of models bigger than 15% of the observed SIE..." to "standard deviation greater than 15%"