Abstract The Earth’s energy budget is fundamental to climate, and its accurate simulations in global climate models (GCMs) are essential for climate projections. While studies have validated energy budget at the top of the atmosphere (TOA) and surface, the layer‐wise atmospheric energy budget remains largely unexplored because of limited observations and diagnostic tools. Here, we use satellite observations and reanalysis to quantify radiative heating rate biases in six GCMs in the Atmospheric Modeling Intercomparison Project experiments, and attribute them using radiative heating rate kernels. We find that radiative heating rate biases dominate physical heating rate biases in the extratropics. Middle‐ and lower‐tropospheric biases are primarily contributed by clouds, whereas high‐latitude upper tropospheric biases are mainly caused by air temperature. The cloud‐induced biases are related to lower tropospheric stability and mid‐tropospheric mixed‐phase cloud parameterizations. These findings underline the importance of validating the atmospheric energy budget, not only at the TOA and surface.