Atmosphere-only experiments are widely used to investigate climate feedbacks simulated in more computationally expensive fully-coupled global climate model simulations. We confirm that this remains a valid approach by comparing the radiative feedbacks and forcing between coupled and atmosphere-only simulations for the latest models taking part in the 6th phase of the Coupled Model Intercomparison Project (CMIP6). For cloud feedbacks, we find a better than previously known correspondence between these experiments, which applies even to the response of individual cloud properties (amount, altitude and optical depth), is present at nearly every geographic location, and holds even when considering atmosphere-only simulations of only 1 year duration. In the tropics, the correspondence between the two experiments is better revealed when considering feedbacks stratified by vertical motion rather than by geography, owing to the non-uniform warming pattern in the coupled experiment. For the lapse rate and surface albedo feedbacks, the correspondence between the two experiments is weaker due to the lack of sea-ice changes in the atmosphere-only experiment. For the across-model relationship between 4xCO2 radiative forcing and feedback, we find a different behavior across experiments in CMIP6 than in CMIP5, casting doubt on the physical significance of previous results that highlighted an anti-correlation between the two quantities. Overall, these results confirm the utility of atmosphere-only experiments particularly to study cloud feedbacks, which are the dominant source of inter-model spread in climate sensitivity.