Model intercomparison studies of coupled carbon-climate simulations have the potential to improve our understanding of the processes explaining the pCO₂ drawdown at the Last Glacial Maximum (LGM) and to identify related model biases. Models participating in the Paleoclimate Modelling Intercomparison Project (PMIP) now frequently include the carbon cycle. The ongoing PMIP-carbon project provides the first opportunity to conduct multimodel comparisons of simulated carbon content for the LGM time window. However, such a study remains challenging due to differing implementation of ocean boundary conditions (e.g. bathymetry and coastlines reflecting the low sea level) and to various associated adjustments of biogeochemical variables (i.e. alkalinity, nutrients, dissolved inorganic carbon). After assessing the ocean volume of PMIP models at the pre-industrial and LGM, we investigate the impact of these modelling choices on the simulated carbon at the global scale, using both PMIP-carbon model outputs and sensitivity tests with the iLOVECLIM model. We show that the carbon distribution in reservoirs is significantly affected by the choice of ocean boundary conditions in iLOVECLIM. In particular, our simulations demonstrate a ~250 GtC effect of an alkalinity adjustment on carbon sequestration in the ocean. Finally, we observe that PMIP-carbon models with a freely evolving CO₂ and no additional glacial mechanisms do not simulate the pCO₂ drawdown at the LGM (with concentrations as high as 313, 331 and 315 ppm), especially if they use a low ocean volume. Our findings suggest that great care should be taken on accounting for large bathymetry changes in models including the carbon cycle.