Xiaomi Auto made another major move in technology, officially launching Xiaomi OneVL—a one-step latent space language-visual reasoning framework. The model is the first in the industry to unify multiple technical routes including VLA, world models and latent space reasoning. Building on the powerful reasoning capabilities of the XLA model, it simultaneously improves reasoning speed and accuracy, outperforming explicit CoT in accuracy while matching the speed of "answer-only" latent CoT solutions.

Notably, Xiaomi announced that it will fully open-source the model weights, training code and inference code of Xiaomi OneVL for free use by global developers. Xiaomi stated that this is not only a supplement to XLA technology but also a deep exploration of latent space reasoning capabilities in autonomous driving large models, which will promote technological progress across the entire autonomous driving industry