Changes in the marine reservoir age (MRA) of the surface ocean are important information used for radiocarbon dating of marine sediment cores or archaeological artifacts. MRA changes are expressed relative to the atmosphere, and as such are dependent on the prevailing atmospheric radiocarbon calibration curve. The most recent estimate for evolving global average MRA for latitudes approximately <50° is incorporated into the marine calibration curve Marine20. This curve was directly calculated from the atmospheric Δ14C record, IntCal20, using the carbon cycle box model BICYCLE, taking into account observed changes in the carbon cycle. These simulations did not consider changes in the strength of the Atlantic meridional overturning circulation (AMOC) related to Dansgaard/Oeschger and Heinrich events. A recent study using the successor BICYCLE-SE suggested that abrupt AMOC changes would lead to changes in MRA of less than 100 14C yr in the non-polar surface ocean (about <50°). To better support previous model-based MRA and to further constrain the impact of AMOC changes on MRA, we here assess transient simulations of the last 55 kyr performed by two Earth System Models of Intermediate Complexity (EMICs), LOVECLIM and Bern3D, and compare them to the published BICYCLE-SE box model results and previous output from the Large Scale Geostrophic (LSG) ocean general circulation model (OGCM). The setups within this MRA model intercomparison (MRA-MIP) are not identical, but all models are forced by atmospheric CO2 and Δ14C to have the surface ocean carbon cycle state as close as possible to reconstructions. Simulations with abrupt AMOC reductions during stadials display a rise in MRA in the surface northern Atlantic (>50° N) and the deep Atlantic, for example reaching 300–1250 and 500–1300 14C yr, respectively, during Heinrich stadial 1. We find that the changes in the mean non-polar surface MRA (<50° latitude) during abrupt AMOC changes in LOVECLIM are also in the order of ±100 14C yr, while in Bern3D simulated changes are up to ±200 14C yr. While the models tend to agree that a reduced AMOC leads to lower MRA by about 100–300 14C yr in the low-latitude surface ocean, under some conditions the opposite is found (e.g. simulations with LOVECLIM across Heinrich stadial 1). Spatially resolved results of the models show that changes in surface MRA during stadials depict the general pattern of a radiocarbon bipolar seesaw (older surface water in the high north, younger in the high south and in the Indo-Pacific), in agreement with previously published reconstructions. However, some model-dependent differences remain in the non-polar Atlantic. Throughout the last 50 kyr, the change in the multi-model mean in non-polar MRA of the two EMICs when compared with Marine20 is less than 100 14C yr and within the uncertainties of Marine20. Furthermore, changes in the MRA of the high latitude Southern Ocean (>50° S) are extremely model-dependent and, for much of the period between 18 and 43 kyr BP, the changes in the multi-model mean MRA are larger than the 95 % confidence interval of the non-polar MRA depicted in Marine20. These differences make the construction of a numerical model-based calibration curve for the high latitude Southern Ocean challenging.