Abstract: In order to evaluate the characteristics of the dynamical core of GRAPES (Global/Regional Assimilation and PrEdiction System) and to verify whether the GRAPES dynamical core can be taken as a framework of AGCM, the long time integration has been carried out with GRAPES dynamical core followed by the benchmark test similar to the technique introduced by Held and Suarez. The GRAPES has a finite-difference dynamical core with fully compressible, non-hydrostatic framework and semi-implicit and semi-lagrangian (SISL) time integration scheme using latitude and longitude horizontal coordinate as well as following terrain height vertical coordinate. The computational design uses two simple physical processes to represent physics processes in the model during the time integration. The first process is to relax the model variant, potential temperature, to a prescribed potential temperature field which is a function of latitude and height during the model time integration. Following Held and Suarez, the second one is the momentum drag in the lower troposphere to consume the kinetic energy in the model atmosphere. The GRAPES dynamical core runs with 31 levels in vertical direction and 3 horizontal resolutions:1.25°, 2.5°and 5.0°latitude or longitude. The benchmark tests integrate for 1460 days under each horizontal resolution and the last 1000 days data are analyzed by discarding the previous 460 day results. The statistic results show that the GRAPES dynamical core can reproduce basic features of atmospheric circulation, including reasonably realistic zonal mean temperature and its eddy variance as well as zonal-mean wind and its eddy variance. A single westerly jet is generated with maximum strength of roughly 28 m/s near 45°latitude in the tropopause. And eddy temperature variance shows two middle latitude maxima, one in the low troposphere and the other above tropopause. The GRAPES dynamical core also shows stability for the energy and momentum as well as vertical wind speed during the long term integration although SISL scheme does not observe the conservations. The GRAPES dynamical core has convergence features along with increasing resolution, although 5.0° resolution in latitude and longitude may be too coarse to represent the circulation details. Finally, the sensitivity to non-hydrostatic is discussed. The results indicate that it is feasible using the GRAPES frame as a dynamical core for AGCM and climate investigation. The numerical experiments provide clues and evidences for GRAPES dynamical core improvement also.