We consider the accuracy and precision of COSMIC-2 (C2) radio occultation retrievals of temperature, specific humidity, and precipitable water (PW) in Hurricane Dorian (2019) and the larger-scale tropical and subtropical environment by comparing them with ERA5, MERRA-2, and JRA-55 reanalyses. Because of its high signal-to-noise ratio, at least 70% of the C2 soundings penetrate to within 1 km of the surface over the oceans, even in the extremely moist hurricane atmosphere. We compute biases and standard deviations of differences of C2 temperature and specific humidity with respect to ERA5 and estimate the standard deviation of random errors of C2 using the three-cornered hat (3CH) method. The C2 profiles of mean temperature and specific humidity agree closely with ERA5 above 2 km, with a temperature bias of less than 0.3 K and a specific humidity bias of less than 0.3 g/kg (15%). Below 2 km, C2 shows a dry bias of up to 2 g/kg relative to ERA5, which is likely related to factors other than super-refraction. This dry bias below 2 km, which is present in more than 90% of the profiles, causes a small (1-2%) dry bias in PW. The 3CH estimates of C2 temperature error standard deviations are less than 0.5 K throughout most of the troposphere. The 3CH error estimates of C2 specific humidity are less than 1.0 g/kg (25%). These estimates are similar to those of ERA5 and smaller than those of MERRA-2 and JRA-55. The 3CH estimates of C2 PW random errors are the smallest of all the data sets. Overall, we conclude that C2 provides accurate and precise vertical profiles of temperature, water vapor, and precipitable water in the hurricane and its environment.