Abstract: The sliding temporal correlation (STC) approach is applied to the study of reliable computation time (RCT) for chaotic numeric systems and general oceanic circulation models. The numeric experiments in the Lorenz chaotic system indicate that the maximal prediction time and RCT calculated by STC well agree with those calculated using the classic error approach. This indicates that the STC approach may be applied to this study. Therefore, this approach is applied to examine the RCT of sea surface temperature (SST) simulated by two uncoupled LICOM and NEMO oceanic models and a coupled model using NEMO as its oceanic component. It is found that the mean RCT of SST extends to about six months in the two uncoupled oceanic models. This relatively long RCT is due to the adoption of surface boundary restore conditions that revert the numeric results to the observed values. Using external forcing data from January, low RCT values (under two months) in simulations are located in the equatorial Pacific and Northwest Pacific for the LICOM model. High RCT values for the NEMO model are located in the equatorial Pacific and low RCT values in the off-equatorial Pacific and in the middle latitudes of the Atlantic. Simulations using external forcing data from July show the reverse zonal mean RCT pattern to those from January. In addition, when using the coupled model (NEMO model coupled with other components such as atmosphere), the restore surface boundary condition is not used. The mean annual RCT in the coupled model decreases significantly to around one month. Seasonal variation in RCT is small, and the commonly seasonal mean RCT is about 30-40 days. In spring, a long RCT (over two months) is found mainly in the Southern Hemisphere and in autumn, mainly in the Northern Hemisphere. The RCT of 500 hPa geopotential height in the coupled model is about two weeks, which is close to that simulated by the uncoupled AGCM model. Moreover, the seasonal mean or regional mean RCT is generally in the range of 30-60 days and only in a few small regions, and in specific seasons, the RCT is longer than 80 days. This indicates that in the coupled model, the maximal prediction time is limited to 2-3 months for SST due to computational uncertainty. These average RCTs are shorter than the predictable time obtained by observation data (about eight months). Thus, due to the cask principle, the RCT may be an important reason to restrict the predictable time length of SST in the coupled model.