This study provides a systematic evaluation of Antarctic sea ice simulations in CMIP6 models by integrating satellite observations and reanalysis data. The paper is well-structured, employs rigorous methodology, and presents comprehensive data analysis. It contributes meaningfully to the field of Antarctic sea ice modeling. Below are my detailed review comments:

1. The manuscript mentions GIOMAS and GECCO3 reanalysis products but does not explicitly clarify whether these products assimilate observational data (e.g., sea ice concentration, satellite retrievals).
2. Lines 211–213 mention that the IPSL and EC-Earth3 models produce anomalously thick ice, but there is no explanation of why these models behave differently (e.g., ocean/ice dynamics, resolution, parameterizations). I suggest adding a paragraph comparing key models, such as IPSL vs. CESM2, to highlight how differences in physical schemes (e.g., ridging, thermodynamics) lead to divergent sea ice thickness simulations.
3. Although the paper discusses biases in CMIP6 models when simulating Antarctic sea ice, the analysis of the underlying causes of these biases is somewhat brief. I recommend a more detailed discussion of the sources of model discrepancies, particularly with respect to how different models perform in different regions and the specific impacts these factors have on sea ice simulations.
4. The right column of Figure 2 (SIA anomalies) excludes GECCO3 and GIOMAS, unlike the left and middle columns (SIT/SIV). I suggest either adding the SIA data from GECCO3 and GIOMAS or providing justification for their exclusion (e.g., NSIDC is the primary SIA reference).
5. The analysis stops at 2014, while Antarctic sea ice has shown a significant decline since 2015 (Raphael & Handcock, 2022). It would be helpful to briefly discuss whether CMIP6 models (e.g., SSP scenarios) capture this reversal, even if reanalysis data does not include post-2014 information.
6. The Taylor diagrams in Figure 3 use "satellite products" as the reference, but it is unclear whether Envisat-CryoSat-2 is averaged or used separately. I recommend explicitly stating which dataset is used as the reference for each variable and justifying any averaging.
7. Section 3.4 relies on GIOMAS as the primary reference due to satellite biases in September. However, GIOMAS shows large errors in February (Figure 1c, early maxima). I suggest acknowledging the limitations of GIOMAS and cross-validating key results with GECCO3 and satellite data where possible.
8. Some of the legends are not clear enough. It is recommended to directly mark them in the figures or explain them in detail in the captions.
9. Although the paper proposes directions for model improvements (e.g., improving ice-edge simulation accuracy, enhancing dynamic process representations), these suggestions are rather general and lack specific implementation details.