Substantial model biases are still prominent even in the latest CMIP6 simulations; attributing their causes is defined as one of the three main scientific questions addressed in CMIP6. In this paper, cold temperature biases in the North Pacific subtropics are investigated using simulations from the newly released CMIP6 models, together with other related modeling products. In addition, ocean-only sensitivity experiments are performed to characterize the biases, with a focus on the role of oceanic vertical mixing schemes. Based on the Argo-derived diffusivity, idealized vertical diffusivity fields are designed to mimic the seasonality of vertical mixing in this region, and are employed in ocean-only simulations to test the sensitivity of this cold bias to oceanic vertical mixing. It is demonstrated that the cold temperature biases can be reduced when the mixing strength is enhanced within and beneath the surface boundary layer. Besides, the temperature simulations are rather sensitive to the parameterization of static instability, and the cold biases can be reduced when the vertical diffusivity for convection is increased. These indicate that the cold temperature biases in the North Pacific can be largely attributed to biases in oceanic vertical mixing within ocean-only simulations, which likely contributes to the even larger biases seen in coupled simulations. This study therefore highlights the need for improved oceanic vertical mixing in order to reduce these persistent cold temperature biases seen across several CMIP models.