Great advances have been made in the last decade in quantifying and understanding the spatiotemporal patterns of terrestrial gross primary production (GPP) with ground, atmospheric, and space observations. However, although global GPP estimates exist, each data set relies upon assumptions and none of the available data are based only on measurements. Consequently, there is no consensus on the global total GPP and large uncertainties exist in its benchmarking. The objective of this review is to assess how the different available data sets predict the spatiotemporal patterns of GPP, identify the differences among data sets, and highlight the main advantages/disadvantages of each data set. We compare GPP estimates for the historical period (1990–2009) from two observation‐based data sets (Model Tree Ensemble and Moderate Resolution Imaging Spectroradiometer) to coupled carbon‐climate models and terrestrial carbon cycle models from the Fifth Climate Model Intercomparison Project and TRENDY projects and to a new hybrid data set (CARBONES). Results show a large range in the mean global GPP estimates. The different data sets broadly agree on GPP seasonal cycle in terms of phasing, while there is still discrepancy on the amplitude. For interannual variability (IAV) and trends, there is a clear separation between the observation‐based data that show little IAV and trend, while the process‐based models have large GPP variability and significant trends. These results suggest that there is an urgent need to improve observation‐based data sets and develop carbon cycle modeling with processes that are currently treated either very simplistically to correctly estimate present GPP and better quantify the future uptake of carbon dioxide by the world's vegetation.