<p>Atmospheric CO<sub>2</sub> increased by about 90 ppm across the transition from the Last Glacial Maximum (LGM) to the end of the preindustrial (PI) period. The contribution of changes in land carbon stocks to this increase remains uncertain. Estimates of the PI-LGM difference in land biosphere carbon inventory (∆land) range from −400 to +1,500 GtC, based on upscaling of scarce paleo soil carbon or pollen data. A perhaps more reliable approach infers ∆land from reconstructions of the stable carbon isotope ratio in the ocean and atmosphere assuming isotopic mass balance with recent studies yielding ∆land values of about 300–400 GtC. Surprisingly, however, earlier studies considered a mass balance for the ocean–atmosphere–land biosphere system only. Thereby, these studies neglect carbon exchange with sediments, weathering-burial flux imbalances, and the influence of the deglacial reorganization on the isotopic budgets. We show this neglect to significantly bias low deglacial ∆land in simulations using the Bern3D Earth System Model of Intermediate Complexity v.2.0s. We constrain ∆land to ∼ 850 GtC (median estimate; 450 to 1250 GtC 1<i>σ</i> range) by using reconstructed changes in atmospheric <i>δ</i><sup>13</sup>C, marine <i>δ</i><sup>13</sup>C, deep Pacific carbonate ion concentration, and atmospheric CO<sub>2</sub> as observational targets in a Monte Carlo ensemble with half a million members. Sensitivities of the target variables to changes in individual deglacial carbon cycle processes are established from factorial simulations over the past 21,000 years with the Bern3D model. These are used in the Monte Carlo ensemble and provide forcing–response relationships for future model–model and model–data comparisons. Uncertainties in the estimate of ∆land remain considerable due to model and proxy data uncertainties. Yet, it is likely that ∆land is larger than 450 GtC and highly unlikely that the carbon inventory in the land biosphere was larger for the LGM than during the recent preindustrial period.</p>