In this paper the performance of the global coupled general circulation model (CGCM) ECHO-2, which was integrated for 10 years without the application of flux correction, is described. Although the integration is rather short, strong and weak points of this CGCM can be clearly identified, especially in view of the model’s performance of the annual cycle in the tropical Pacific. The latter is simulated with more success relative to the earlier version, ECHO-1. A better representation of the low-level stratus clouds in the atmosphere model associated with a reduction in the shortwave radiative flux at the air–sea interface improved the coupled model’s performance in the southeastern tropical oceans, with a strongly reduced warm bias in these regions. Modifications in the atmospheric convection scheme also eliminated the AGCM’s tendency to simulate a double ITCZ, and this behavior is maintained in the CGCM simulation. Finally, a new numerical scheme for active tracer advection in the ocean model strongly reduced the numerical mixing, which seems to enhance considerably the level of interannual variability in the equatorial Pacific. One weak point is an overall cold bias in the Tropics and midlatitudes, which typically amounts to 1°C in open ocean regions. Another weak point is the still too strong equatorial cold tongue, which penetrates too far into the western equatorial Pacific. Although this model deficiency is not as pronounced as in ECHO-1, the too strong cold tongue reduces the level of interannual rainfall variability in the western and central equatorial Pacific. Finally, the interannual fluctuations in equatorial Pacific sea surface temperatures (SSTs) are too equatorially trapped, a problem that is also found in “ocean-only” simulations. Overall, however, the authors believe that the ECHO-2 CGCM has been considerably improved relative to ECHO-1.