We use 148 airborne vertical profiles of CO2 for frontal cases from the summer 2016 Atmospheric Carbon and Transport—America (ACT-America) campaign to evaluate the skill of 10 global CO2 in situ inversion models from the version 7 Orbiting Carbon Observatory–2 (OCO-2) Model Intercomparison Project (MIP). Model errors (model posterior—observed CO2 dry air mole fractions) were categorized by region (Mid-Atlantic, Midwest, and South), frontal sector (warm or cold), and transport model (predominantly Tracer Model 5 (TM5) and Goddard Earth Observing System—Chemistry [GEOS-Chem]). All inversions assimilated the same CO2 observations. Overall, the median inversion profiles reproduce the general structures of the observations (enhanced/depleted low-level CO2 in warm/cold sectors), but 1) they underestimate the magnitude of the warm/cold sector mole fraction difference, and 2) the spread among individual inversions can be quite large (>5 ppm). Uniquely in the Mid-Atlantic, inversion biases segregated according to atmospheric transport model, where TM5 inversions biases were −3 to −4 ppm in warm sectors, while those of GEOS-Chem were +2 to +3 ppm in cold sectors. The large spread among the mean posterior CO2 profiles is not explained by the different atmospheric transport models. These results show that the inversion systems themselves are the dominant cause of this spread, and that the aircraft campaign data are clearly able to identify these large biases. Future controlled experiments should identify which inversions best reproduce mid-latitude CO2 mole fractions, and how inversion system components are linked to system performance.