Xiaomi's AI team open-sources a large-scale autonomous driving model, Xiaomi OneVL.

Xiaomi's Embodied Intelligence Team has open-sourced the Xiaomi OneVL one-step latent space language and visual inference framework. The model is the first in the industry to unify VLA, world model, and latent space inference, possessing both powerful inference capabilities and high-speed response. Its accuracy surpasses explicit CoT, and its speed matches latent space CoT solutions. It achieves state-of-the-art (SOTA) performance on multiple benchmarks, including ROADWork, Impromptu, and Alpamayo-R1, providing interpretability in both language and vision dimensions.