Another Chinese lab has put its weights online. Xiaomi’s MiMo-V2.6 line now ships free for anyone to download, with a large Pro model and a cheaper Flash sibling, both able to read images, audio and video as well as text.
The company says Pro scored 46 on the Artificial Analysis Intelligence Index, ahead of Kimi K3 at 44 and Zhipu’s GLM-5.3 at 45, making it the highest-ranked open-weight model on that measure. Closed frontier systems scored 53.
Both models accept text, image, video and audio with one million-token context windows, a design aimed at agentic workloads such as computer use, where a single model loop reads a task, interprets a screenshot and decides what to do next. Xiaomi also shipped a distilled 9B model and released reinforcement learning research.
Cheap to train, fast to run
The two models completed 30 reinforcement learning steps in under six days, using roughly 750,000 trajectories at reported costs of $2.62M for Pro and $850,000 for Flash. A Pro-UltraSpeed variant promises output up to twenty times faster at the same quality.
Xiaomi presents the release as a bid for developer mindshare that Chinese labs have been winning from Western counterparts, giving builders frontier-adjacent capability without a large API bill.
