To evaluate the ability of CMIP6 Earth System Models to simulate net biome production (NBP) changes caused by natural and man‐made forcing, the simulated NBP during the period 1901–2010 at global, hemispheric and regional scales was compared with the estimated NBP of the TRENDY ensemble. The annual NBP, globally and over the Northern Hemisphere and tropics, from the CMIP6 ensemble reflected well the TRENDY ensemble annual values from 1901 to 2010, and the correlation coefficient passed the 0.05 significance level. The annual NBP estimated by CMIP6 over the globe, Northern Hemisphere and tropics was 0.20 ± 0.46, 0.25 ± 0.34, and −0.02 ± 0.14 Pg C yr−1, respectively; while from the TRENDY ensemble it was 1.04 ± 0.41, 0.55 ± 0.24, and 0.43 ± 0.20 Pg C yr−1, respectively. The main weaknesses of the CMIP6 multi‐model ensemble simulation was the lower strength of NBP, especially at low latitudes. The simulated results showed higher soil respiration estimated by CMIP6 for these regions, especially in the Amazon Basin, than that by the TRENDY ensemble, which produced the greater NPB estimated by TRENDY. The CMIP6 ensemble exaggerated the increasing trend of carbon sinks between 30° and 60°N in the Northern Hemisphere after 1961, due to the stronger positive response of NPP (net primary production) in the high‐latitude Northern Hemisphere to global warming. The NBP during 1901–2010 could not be reproduced well by the CMIP6 ensemble in most sub‐regions, as evidenced by the failure to pass the statistical test at the 5% significance level in 70% of the sub‐regions in terms of the correlation coefficients with the TRENDY ensemble. In eastern Asia, for example, NBP was poorly estimated by the CMIP6 ensemble, as evidenced by negative correlation coefficients with the TRENDY ensemble.