On the afternoon of August 24, Xiaomi convened a technical communication session centered around its Xuanjie chips. During this event, Lei Jun shared comprehensive details about two prototype devices that are powered by the upcoming generation of Xuanjie chips.
The Xuanjie O100 prototype stands out as a high-speed AI demonstration terminal tailored for edge-side large models. It incorporates a dual-chip collaborative architecture, does away with the conventional mobile phone imaging module, revamps the motherboard design, and is outfitted with an active air cooling mechanism. The edge-side model embedded within it attains an impressive measured inference speed of up to 295 Tokens per second.
Another notable prototype is the Xiaomi AI Cube, which serves as a personal AI supercomputing terminal. Its casing is crafted from aviation aluminum and features a multitude of heat dissipation holes, facilitating a sustained high-performance output of 150 watts. This device integrates three chips, boasts 80GB of unified memory, accommodates the deployment of models with diverse parameter counts, and permits seamless switching between fast and slow systems.
Lei Jun further disclosed that the Xuanjie O3 chip will be a key feature in the Xiaomi 18 Fold. Meanwhile, the Xuanjie O100 and D100 are projected to make their commercial debut in 2027.
