The Pacific–Japan (PJ) teleconnection pattern, a dominant mode of atmospheric variability over the western North Pacific during boreal summer, is pivotal in shaping regional climate dynamics. Despite its important implications, accurately predicting the PJ pattern remains challenging due to inherent model biases and uncertainties. This study delves into the impact of model biases on the prediction skill of the PJ pattern and evaluates its predictability using outputs from three operational seasonal forecast models. Our findings elucidate that the spatial structure of the PJ pattern simulated by models introduces substantial diversities in prediction skills. By discerning the variance in PJ teleconnection simulation among models, we unveil the high predictability of the PJ pattern, showcasing its capability for accurate forecasts up to 3 months in advance within the current seasonal forecast models. The predictability of the PJ pattern stems from concurrent El Niño–Southern Oscillation related sea surface temperature anomalies and its corresponding atmospheric teleconnection processes. Our research underscores the necessity of accounting for model biases in predicting the PJ pattern, and the potential for bolstering seasonal prediction skill through targeted mitigation of these biases.