Abstract. Soil organic carbon (SOC) is the largest carbon pool in terrestrial ecosystems and may play a key role in biospheric feedbacks with elevated atmospheric carbon dioxide (CO₂) in a warmer future world. We examined the simulation results of seven terrestrial biome models when forced with climate projections from four representative-concentration-pathways (RCPs)-based atmospheric concentration scenarios. The goal was to specify calculated uncertainty in global SOC stock projections from global and regional perspectives and give insight to the improvement of SOC-relevant processes in biome models. SOC stocks among the biome models varied from 1090 to 2650 Pg C even in historical periods (ca. 2000). In a higher forcing scenario (i.e., RCP8.5), inconsistent estimates of impact on the total SOC (2099–2000) were obtained from different biome model simulations, ranging from a net sink of 347 Pg C to a net source of 122 Pg C. In all models, the increasing atmospheric CO₂ concentration in the RCP8.5 scenario considerably contributed to carbon accumulation in SOC. However, magnitudes varied from 93 to 264 Pg C by the end of the 21st century across biome models. Using the time-series data of total global SOC simulated by each biome model, we analyzed the sensitivity of the global SOC stock to global mean temperature and global precipitation anomalies (ΔT and ΔP respectively) in each biome model using a state-space model. This analysis suggests that ΔT explained global SOC stock changes in most models with a resolution of 1–2 °C, and the magnitude of global SOC decomposition from a 2 °C rise ranged from almost 0 to 3.53 Pg C yr⁻¹ among the biome models. However, ΔP had a negligible impact on change in the global SOC changes. Spatial heterogeneity was evident and inconsistent among the biome models, especially in boreal to arctic regions. Our study reveals considerable climate uncertainty in SOC decomposition responses to climate and CO₂ change among biome models. Further research is required to improve our ability to estimate biospheric feedbacks through both SOC-relevant and vegetation-relevant processes.