Earth’s climate naturally fluctuates on intraseasonal to interdecadal timescales, partly due to processes intrinsic to the climate system. Simulation ensembles can illuminate the role of this internal variability and better constrain climate projections. We analyze the first three statistical moments of precipitation and sea surface temperature in the eastern equatorial Pacific simulated by climate models from the 6th Coupled Model Intercomparison Project (CMIP6) to highlight how internal variability reduces our capacity to robustly evaluate climate models. We show that, as expected according to the statistical theory, the precision of the ensemble mean is related to the ensemble standard deviation and the square root of the ensemble size. We also show that increasing the epoch length used to compute the statistic has a similar influence as increasing the ensemble size, but with larger inter-model difference. As the precision of the ensemble mean is, on average, similar when computed using pre-Industrial control and historical runs, we show how control runs can be used to estimate the ensemble size for historical runs and provide examples of methods to decide on the ensemble size. In addition, the correspondence between control and historical precisions allows to estimate precision in the latter when a large ensemble is not available.