Abstract A benchmark calculation is designed to compare the climate and climate sensitivity of atmospheric general circulation models (AGCMs). The experimental setup basically follows that of the aquaplanet experiment (APE) proposed by Neale and Hoskins, but a simple mixed layer ocean is embedded to enable air–sea coupling and the prediction of surface temperature. In calculations with several AGCMs, this idealization produces very strong zonal-mean flow and exaggerated ITCZ strength, but the model simulations remain sufficiently realistic to justify the use of this framework in isolating key differences between models. Because surface temperatures are free to respond to model differences, the simulation of the cloud distribution, especially in the subtropics, affects many other aspects of the simulations. The analysis of the simulated tropical transients highlights the importance of convection inhibition and air–sea coupling as affected by the depth of the mixed layer. These preliminary comparisons demonstrate that this idealized benchmark provides a discriminating framework for understanding the implications of differing physics parameterization in AGCMs.