Abstract. Most state-of-the-art climate models include a coarsely resolved oceanic component, which hardly captures detailed dynamics, whereas eddy-permitting and eddy-resolving simulations are developed to reproduce the observed ocean. In this study, an eddy-permitting and a coarse resolution numerical experiment are conducted to simulate the global ocean state for the period of the Last Glacial Maximum (LGM, ~26 500 to 19 000 yr ago) and to investigate the improvements due to taking into account the smaller spatial scales. The ocean state from each simulation is confronted with a data set from the Multiproxy Approach for the Reconstruction of the Glacial Ocean (MARGO) sea surface temperatures (SSTs), some reconstructions of the palaeo-circulations and a number of sea-ice reconstructions. The western boundary currents and the Southern Ocean dynamics are better resolved in the high-resolution experiment than in the coarse simulation, but, although these more detailed SST structures yield a locally improved consistency between model predictions and proxies, they do not contribute significantly to the global statistical score. The SSTs in the tropical coastal upwelling zones are also not significantly improved by the eddy-permitting regime. The models perform in the mid-latitudes but as in the majority of the Paleoclimate Modelling Intercomparison Project simulations, the modelled sea-ice conditions are inconsistent with the palaeo-reconstructions. The effects of observation locations on the comparison between observed and simulated SST suggest that more sediment cores may be required to draw reliable conclusions about the improvements introduced by the high resolution model for reproducing the global SSTs. One has to be careful with the interpretation of the deep ocean state which has not reached statistical equilibrium in our simulations. However, the results indicate that the meridional overturning circulations are different between the two regimes, suggesting that the model parametrizations might also play a key role for simulating past climate states.