In February 2023, Antarctic sea ice extent reached a record minimum, yet the mechanisms driving this event remain debated.<br />Here we use a coupled ocean–sea ice model (NEMO-SI3) forced by two different atmospheric reanalyses, ERA5 and JRA-55-<br />do, to investigate the drivers of the 2023 minimum and assess the sensitivity of inferred mechanisms to reanalysis choice. Both<br />simulations capture exceptionally low February sea ice extent during the 2022/23 melt season, though neither atmospheric5<br />reanalysis reproduces 2023 as a record minimum, with JRA-55-do showing closer agreement with observations. Substantial<br />regional differences emerge between the two simulations in February 2023: JRA-55-do produces greater sea ice coverage in<br />the Amundsen–Bellingshausen and Ross sectors, while ERA5 yields higher concentrations in the Weddell and Indian Ocean<br />sectors. We further investigate the atmospheric differences between the atmospheric reanalyses - differences in downwelling<br />longwave radiation are the strongest disparity in October, whereas shortwave radiation is the primary disparity in February-10<br />suggesting a role for these atmospheric differences in the differing sea ice outcome between the two simulations. Much of the<br />existing literature attributes the 2023 minimum to mechanisms inferred from a single reanalysis, most commonly ERA5. Our<br />results demonstrate that reanalysis choice produces markedly different regional sea ice responses, with implications for the<br />interpretation of both thermodynamic and dynamical drivers. We therefore recommend caution when diagnosing real-world<br />sea ice events from reanalysis-forced simulations and that multiple reanalysis products could be employed to improve the15<br />robustness of inferred mechanisms