Abstract Reliable estimates of externality costs—such as the costs arising from premature mortality due to exposure to fine particulate matter (PM_2.5)—are critical for policy analysis. To facilitate broader analysis, several datasets of the social costs of air quality have been produced by a set of reduced-complexity models (RCMs). It is much easier to use the tabulated marginal costs derived from RCMs than it is to run ‘state-of-the-science’ chemical transport models (CTMs). However, the differences between these datasets have not been systematically examined, leaving analysts with no guidance on how and when these differences matter. Here, we compare per-tonne marginal costs from ground level and elevated emission sources for each county in the United States for sulfur dioxide (SO₂), nitrogen oxides (NO_x), ammonia (NH₃) and inert primary PM_2.5 from three RCMs: Air Pollution Emission Experiments and Policy (AP2), Estimating Air pollution Social Impacts Using Regression (EASIUR) and the Intervention Model for Air Pollution (InMAP). National emission-weighted average damages vary among models by approximately 21%, 31%, 28% and 12% for inert primary PM_2.5, SO₂, NO_x and NH₃ emissions, respectively, for ground-level sources. For elevated sources, emission-weighted damages vary by approximately 42%, 26%, 42% and 20% for inert primary PM_2.5, SO₂, NO_x and NH₃ emissions, respectively. Despite fundamental structural differences, the three models predict marginal costs that are within the same order of magnitude. That different and independent methods have converged on similar results bolsters confidence in the RCMs. Policy analyzes of national-level air quality policies that sum over pollutants and geographical locations are often robust to these differences, although the differences may matter for more source- or location-specific analyzes. Overall, the loss of fidelity caused by using RCMs and their social cost datasets in place of CTMs is modest.