A fundamental aspect of marine benthic foraminiferal δ18O records of the last one million years is their “saw-tooth” pattern, characterized by gradual cooling to full glacial conditions followed by very rapid glacial terminations. Understanding the mechanism of this pattern is crucial for understanding ice age dynamics. We investigated the climatic trend within each glacial of the last 880 kyr by comparing the signal of the East Asian winter monsoon (EAWM) with global terrestrial and marine paleoclimate records. Our new grain size record from the Huining loess-paleosol section in the western Chinese Loess Plateau, together with published climate records from the Yimaguan and Luochuan sections in the central Chinese Loess Plateau, demonstrates a strengthening trend of the EAWM within most glacials, which largely mirrors the “saw-tooth” pattern of the benthic foraminiferal δ18O record. However, the EAWM record during Marine Isotope Stage (MIS) 8 shows an unusual weakening trend which is the exception over the last 880 kyr. Based on the linkages between Arctic ice sheets, the Siberian High Pressure system, and the EAWM, we propose that the weakening trend of the EAWM throughout MIS 8 reflects the overall shrinking of Arctic ice sheets towards the glacial termination, in contrast to the increasing trend of the benthic foraminiferal δ18O values. This conclusion is supported by a compilation of global paleoclimate records suggesting that MIS 8 was a unique glacial period over the last 900 kyr, characterized by a mild climate in middle to high northern latitudes and a cold climate in high southern latitudes, during late MIS 8. The inferred changes in Arctic ice sheets would alone cause a decreasing trend in benthic foraminiferal δ18O values during MIS 8, which is the opposite to the observed trend. We suggest that an increased ice volume in high southern latitudes led to the cooling of Antarctic Bottom Water and hence to a decrease in ocean bottom water temperature with an effect to increase benthic foraminiferal δ18O during MIS 8 as observed. Our results highlight the complexity of using the benthic δ18O signal alone as a paleoclimate proxy, and that MIS 8 is an important time window for better understanding glacial dynamics driven by changes in the polar regions of both hemispheres.