Carbon dioxide (CO2) fluxes by different methods vary largely at global, regional and local scales. The net CO2 fluxes by three bottom-up methods (tower observation (TWR), biogeochemical models (GTM), and a data-driven model (SVR)), and an ensemble of atmospheric inversions (top-down method, INV) are compared in Yakutsk, Siberia for 2004–2013. The region is characterized by highly homogeneous larch forest on a flat terrain. The ecosystem around Yakutsk shows a net sink of CO2 by all the methods (means during 2004–2007 were 10.9 g C m−2 month−1 by TWR, 4.28gCm−2 month−1 by GTM, 5.62 g C m−2 month−1 and 0.863 g C m−2 month−1 by SVR at two different scales, and 4.89 g C m−2 month−1 by INV). Absorption in summer (June–August) was smaller by three bottom-up methods (ranged from 88.1 to 191.8 g C m−2 month−1) than the top-down method (223.6 g C m−2 month−1). Thus the peak-to-trough amplitude of the seasonal cycle is greater for the inverse models than bottom-up methods. The monthly-mean seasonal cycles agree among the four methods within the range of inter-model variations. The interannual variability estimated by an ensemble of inverse models and a site-scale data-driven model (the max-min range was 35.8 g C m−2 month−1and 34.2 g C m−2 month−1) is more similar to that of the tower observation (42.4 g C m−2 month−1) than those by the biogeochemical models and the large-scale data-driven model (9.5 g C m−2 month−1 and 1.45 g C m−2 month−1). The inverse models and tower observations captured a reduction in CO2 uptake after 2008 due to unusual waterlogging.