The rate of carbon sequestration in soil represents a substantial source of uncertainty in projections of future climate change under rising atmospheric carbon dioxide concentrations. Earth system models (ESMs) do not agree on the sign, let alone the magnitude, of the projected 21st-century soil carbon sink. Here we evaluate the radioactive carbon isotope ¹⁴C (radiocarbon) as a constraint on rates of soil cycling in terrestrial ecosystem models. We compare the Community Land Model (CLM) and the Energy Exascale Earth System Model (E3SM) Land Model (ELM) to a global database of soil radiocarbon observations. These are among the few land surface models in ESMs that explicitly represent a radiocarbon tracer, and we found that both models tend to produce radiocarbon ages with a young bias, particularly at depth. This young bias may be symptomatic of unrealistically fast cycling of soil carbon though these models, which could lead to a low bias in the stocks of soil organic carbon or an overestimate of the short-term soil carbon sink. We also use idealized representations of soil carbon models to illustrate the importance of using soil carbon stocks as a constraint when correcting radiocarbon age biases. We further evaluate the tradeoffs between these multiple constraints when calibrating soil carbon parameters in ELM. These results demonstrate the benefit of carbon stocks and radiocarbon ages as multiple constraints when assessing the rate of soil carbon sequestration.