Xiaomi unveils Xuanjie O100 prototype and AI Cube with 295 tokens per second on-device AI speed

Xiaomi's CEO Lei Jun revealed two prototypes powered by the new Xuanjie chips. The Xuanjie O100 prototype is built for on-device AI, using a dual-chip setup with O3 and O100, removing the camera module and adding active cooling that supports up to 10W. It runs Xiaomi's MiMo on-device model, hitting 295 tokens per second in tests. The AI Cube prototype has an aluminum body with 33,874 CNC cooling holes, delivers 150W sustained performance, and packs O3, O100, and D100 chips with 80GB unified memory. It can run a 120B-parameter model alongside a 3B model, making it suitable for complex coding tasks. The O3 chip will debut in the Xiaomi 18 Fold, while the other two chips are slated for commercial use in 2027.