Xiaomi founder, chairman and CEO Lei Jun published an article on September 18 titled "Smoother Conversations, Smarter Car Control — A Technical Look Behind Super Xiao Ai." The Expert Mode of Xiaomi HyperOS Super Xiao Ai has received core technical upgrades centered on two capabilities — a fast/slow system and full-duplex conversation — enabling in-car task orchestration in which "one sentence gets it done."
On the fast/slow system, the fast path handles high-frequency, simple requests such as "turn on the air conditioning" or "navigate to the office," while the slow path takes on complex requests including multi-waypoint navigation, conditional filtering and long-horizon task orchestration. Built on the Xiaomi MiMo model, Super Xiao Ai switches from think mode to no-think mode, cutting average model latency by about 35%; at the same time, RAG-Planning is introduced to retrieve similar problem-solving approaches from a knowledge base as examples, bringing no-think mode up to the accuracy of think mode and shortening end-to-end response times by as much as 3 to 6 seconds.
Super Xiao Ai technical upgrade set to improve user experience
On the full-duplex conversation front, Super Xiao Ai brings rejection, end-of-turn detection and interruption handling into a single control module for unified decision-making, shifting from "listen, then speak" to "speak while listening" so users can interject at any time. For multi-turn dialogue, a three-tier Session compression scheme reduces the context Token volume by 60% to 86%. For interruption handling, the system distinguishes between broadcast interruption, task interruption, misinterruption and fast-stealing/slow-resend cases, with the core criterion being whether the output is an "action" or "information": if one step is skipped, finish the task; if part of the speech is missed, end the turn there.
The Expert version of Super Xiao Ai is currently available on Xiaomi Pengcheng SkyNomad vehicles, and Super Xiao Ai on other car models and devices will receive the upgrade in the future.