This article describes a new capability for high-precision 14C measurement of CO2 from air at the Rafter Radiocarbon Laboratory, GNS Science, New Zealand. We evaluate the short-term within-wheel repeatability and long-term between-wheel repeatability from measurements of multiple aliquots of control materials sourced from whole air. Samples are typically measured to 650,000 14C counts, providing a nominal accelerator mass spectrometry (AMS) statistical uncertainty of 1.3‰. No additional uncertainty is required to explain the within-wheel variability. An additional uncertainty factor is needed to explain the long-term repeatability spanning multiple measurement wheels, bringing the overall repeatability to 1.8‰, comparable to other laboratories measuring air materials to high precision. This additional uncertainty factor appears to be due to variability in the measured 14C content of OxI primary standard targets, likely from the combustion process. We observe an offset of 1.4‰ in our samples relative to those measured by the University of Colorado INSTAAR, comparable to interlaboratory offsets observed in recent intercomparison exercises.