A new approach to validating climate models is introduced. First, four data series are combined into one data product using a hierarchical statistical approach. The resulting estimates of trend and residual spectral density functions are provided with simultaneous error bands. A similar decomposition is made for CMIP6 historical simulations, and the trend and spectral density estimates are plotted on top of the data-based bands. We find that 4 out of 58 models have trends fitting inside the bands, and that 6 out of 58 have spectral density functions agreeing with the data. The difference in the average temperatures from 1995--2014 relative to 1880--1899 is more often smaller in the model runs than in the data.