Realistically representing the present-day characteristics of extreme precipitation has been a challenge for global climate models, which is due in part to deficiencies in model resolution and physics, but is also due to a lack of consistency in gridded observations. In this study, we use three observation datasets, including gridded rain gauge and satellite data, to assess historical simulations from sixteen Coupled Model Intercomparison Project Phase 6 (CMIP6) models. We separately evaluate summer and winter precipitation over the United States (US) with a comprehensive set of extreme precipitation indices, including an assessment of precipitation frequency, intensity and spatial structure. The observations exhibit significant differences in their estimates of area-average intensity distributions and spatial patterns of the mean and extremes of precipitation over the US. In general, the CMIP6 multi-model mean performs better than most individual models at capturing daily precipitation distributions and extreme precipitation indices, particularly in comparison to gauge-based data. Also, the representation of the extreme precipitation indices by the CMIP6 models is better in the summer than winter. Although the "standard" horizontal-resolution can vary significantly across CMIP6 models, from ~0.7˚ to ~2.8˚, we find that resolution is not a good indicator of model performance. Overall, our results highlight common biases in CMIP6 models and demonstrate that no single model is consistently the most reliable across all indices.