Stochastic physics (SP) schemes provide a more realistic representation of the unresolved scales in global circulation models by improving both mean climate and climate variability. We study the impact of including an SP scheme in the atmospheric component of EC-Earth on the simulated climate. In particular, we analyze the evolution of the sea-ice extent in the Arctic during long-term simulations covering the historical and future periods. The experiments consist of coupled climate simulations in which three ensemble members constitute the control runs (base) and three ensemble members include stochastic physics (stoc). For the latter, the Stochastically Perturbed Parametrization Tendencies (SPPT) scheme is incorporated in the atmospheric component of EC-Earth. The original experiments, that are part of the SPHINX project, span from 1850 to 2100. We have additionally extended each simulation for 60 years; the future scenario corresponds to the CMIP5 RCP8.5 set up. We compare both sets of experiments to investigate the climate response to a perturbed atmosphere. The simulated Arctic sea-ice extent in September and March display an overall decrease. The sea-ice loss results faster in the base experiments than in the stoc ones. The model simulates an abrupt sea-ice loss in March that takes place about 10 years earlier in the base experiments than in the stoc ones. The evolution of the global annual mean surface air temperature differs if the SP is on or off. Curves start separating by the second half of the 20th century; reach the maximum difference in 2100 and become almost indistinguishable around 2110. Our results suggest that the transient climate sensitivity is lower when the SP is on than when it is off during the 21st century. However, the opposite occurs when the Arctic is free of sea ice along the whole year. This behavior might be the consequence of the asymmetric effect of stochastic perturbations on the process of condensation. We are now investigating the differences in the albedo and cloud feedbacks between both sets of experiments and the possible influence of the mean state on the model climate sensitivity.