88Z AI, 14 May — Xiaomi's Embodied AI team has open-sourced Xiaomi OneVL, a one-step latent space language-vision reasoning framework. The model is the first in the industry to unify VLA, world models, and latent space reasoning, combining strong reasoning capabilities with high-speed response. Its accuracy surpasses explicit CoT, while its speed aligns with latent space CoT solutions. It achieves SOTA on multiple benchmarks including ROADWork, Impromptu, and Alpamayo-R1, offering dual-dimensional interpretabili
At a glance
- First to unify VLA, world models, and latent space reasoning, surpassing explicit CoT in accuracy with high speed
- Achieves SOTA on benchmarks like ROADWork, offering dual-dimensional interpretability in language and vision
The update
Xiaomi's Embodied AI team has officially open-sourced Xiaomi OneVL, an innovative one-step latent space language-vision reasoning framework. This model marks the industry's first unification of Vision-Language-Action (VLA), world models, and latent space reasoning architectures. With this breakthrough, OneVL delivers robust logical reasoning capabilities while maintaining high-speed responsiveness. Its reasoning accuracy surpasses traditional explicit Chain-of-Thought methods, while its operational speed aligns with latent space Chain-of-Thought solutions. The model has achieved state-of-the-art performance on multiple authoritative benchmarks, including ROADWork, Impromptu, and Alpamayo-R1. Furthermore, it provides interpretability across both language and vision dimensions, enhancing the transparency of model decisions.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- IT之家Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.