We present a detailed comparison of geopotential height fields between the Coupled Model Inter-Comparison Project phase 5 (CMIP5) models and satellite observations from GPS radio occultation (RO). Our comparison focuses on the annual mean, seasonal cycle, and interannual variability of 200 hPa geopotential height in the years 2002–2008. Using a wide sample of atmosphere-only model runs (AMIP) from the CMIP5 archive, we find that most models agree well with the observations and weather reanalyses in the tropics in both the annual means and interannual variabilities. However, the agreement is poor over the extratropics with the largest model spreads in the high latitudes and the largest bias in the southern middle to high latitudes that persist all seasons. The models also show excessive seasonal variability over the Northern midlatitude land areas as well as the Southern Ocean but insufficient variability over the tropics and Antarctica. While the underlying causes for the model discrepancies require further analyses, this study demonstrates that global observations from GPS RO provide accurate benchmark-quality measurements in the upper troposphere and lower stratosphere through which biases in climate models as well as weather reanalyses can be identified.