Significant biases have persisted throughout the latest phases of the Coupled Model Intercomparison Project (CMIP), with important implications for climate science and policy. Here the systematic errors in mean and variability of precipitation (P), total column water vapor (W) and sea surface temperature (SST) were analyzed over the tropical oceans for ten CMIP phase 6 (CMIP6) models. This is a particularly challenging region for climate simulations. A process-based error analysis was employed, grounded by state-of-the-art satellite observations and reanalysis datasets. The results revealed a generalized overestimation of P (up to + 1.5mm/day) and underestimation of P variability (up to -54%), largely due to issues in simulating the intensity and extension of deep convection. The P variability errors were further associated with uncertainties in precipitation sensitivity to SST and W, El-Niño Southern Oscillation (ENSO) variability, and atmospheric longwave and shortwave radiative interactions. The models showed both positive and negative W biases (up to 4.5mm), resulting from differences in water vapor feedback and tropical subsidence. Most models overestimated W variability (up to + 103%) due to misrepresentation of the water vapor feedback, tropical deep convection, and ENSO variability. Concerning SST, the biases were relatively low (below ±1.2K), but the variability was generally overestimated (up to + 110%). The SST errors showed multiple connections, including to water vapor and Bjerknes feedbacks, atmospheric radiative absorption and cooling, and ENSO. Overall, the present results support the importance of multivariate model evaluation frameworks, and the critical need for significant improvements in simulation of deep tropical convection and atmospheric-radiative interactions.