Xiaomi has unveiled its first AI acceleration chip specifically designed for on-device large models—the Xuanjie O100. This chip employs the industry's first 6nm wafer-level vertical stacking packaging and hybrid bonding process, achieving a bonding pitch of 1.4 microns and boasting an ultra-high bandwidth of 1.22TB/s, which is 16 times that of traditional flagship smartphones, with on-device inference speeds reaching up to 330 TPS.