To understand the mechanisms of long-term climate and carbon cycle feedback with anthropogenic impact, past simulations (1850–2005) and projection experiments (2006 to 2100) were conducted using a new Earth system model named “MIROC-ESM”, forced by four representative concentration pathway (RCP) scenarios that describe how greenhouse gases (GHGs), aerosols, and land-use will develop in the future. From these projections, temperature rise from 1850 to 2100 ranged from 2.4 K in the RCP2.6 scenario to 6.2 K in the RCP8.5 scenario. We found that there are discrepancies between the RCPs and the estimates of our model in both allowable fossil fuel and land-use change emissions. The former showed systematic discrepancies likely due to strong positive feedbacks in the model, but the latter did not. The likely reason for the difference in land-use emissions is the modeling of land-use change processes or definitions for the emission. Climate response to the increase of atmospheric carbon showed large variation among scenarios, strongly affected by ocean heat uptake efficiency that could depend on the rate of atmospheric CO₂ increase in each scenario. Large variation between scenarios was also found in carbon cycle sensitivity measured by cumulative airborne fraction. The variation in carbon cycle sensitivity may be attributable to the dependence of concentration-carbon feedback on the rate of atmospheric CO₂ increase. The earth system would show a similar response to emitted carbon during the 21st century if the difference of ocean heat uptake efficiency between scenarios were small. The earth system responds to RCP6.0 with less sensitivity to emitted carbon when compared with other scenarios because of high-efficiency carbon uptakes by land and ocean ecosystems. In contrast, RCP2.6 showed high sensitivity of the earth system to carbon emission, and apparently showed different behavior from other scenarios due to early reduction of GHGs.