Benchmarking has been widely used to assess the ability of climate models to capture the spatial and temporal variability of observations during the historical era. For the carbon cycle and terrestrial ecosystems, the design and development of an open-source community platform has been an important goal as part of the International Land Model Benchmarking (ILAMB) project. Here we developed a new benchmarking software system that enables the user to specify the models, benchmarks, and scoring metrics, so that results can be tailored to specific model intercomparison projects. Evaluation data sets included soil and aboveground carbon stocks, fluxes of energy, carbon and water, burned area, leaf area, and climate forcing and response variables. We used this system to evaluate simulations from the 5^thPhase of the Coupled Model Intercomparison Project (CMIP5) with prognostic atmospheric carbon dioxide levels over the period from 1850 to 2005 (i.e., esmHistorical simulations archived on the Earth System Grid Federation). We found that the multi-model ensemble had a high bias in incoming solar radiation across Asia, likely as a consequence of incomplete representation of aerosol effects in this region, and in South America, primarily as a consequence of a low bias in mean annual precipitation. The reduced precipitation in South America had a larger influence on gross primary production than the high bias in incoming light, and as a consequence gross primary production had a low bias relative to the observations. Although model to model variations were large, the multi-model mean had a positive bias in atmospheric carbon dioxide that has been attributed in past work to weak ocean uptake of fossil emissions. In mid latitudes of the northern hemisphere, most models overestimate latent heat fluxes in the early part of the growing season, and underestimate these fluxes in mid-summer and early fall, whereas sensible heat fluxes show the opposite trend.