# Unitree Open-Sources General Humanoid Foundation Model UnifoLM-WLA-1.0

- Category: Flash (快讯)
- Published: 2026-09-15T11:58:00.000Z
- Source date: 2026-09-15
- HTML: https://www.88z.ai/en/news/flash-20260915-63a0581dfa520dda
- Markdown: https://www.88z.ai/en/news/flash-20260915-63a0581dfa520dda.md
- Other language: https://www.88z.ai/news/flash-20260915-63a0581dfa520dda.md

Unitree has fully open-sourced its new general humanoid foundation model, UnifoLM-WLA-1.0, releasing the code, model weights, and datasets simultaneously. Trained on large-scale multimodal perception and interaction-centric world modeling using approximately 2,500 hours of real-robot data, the model achieves leading performance in embodied reasoning benchmarks, rivaling top closed-source models. It enables a single model to manage 64 tasks, covering both desktop manipulation and whole-body mobil

## At a glance

- UnifoLM-WLA-1.0 is fully open-source with code, model, and datasets, trained on 2,500 hours of real-robot data.
- A single model manages 64 tasks with cross-task and cross-end-effector generalization, rivaling top closed-source models.

## The update

Unitree has officially launched and fully open-sourced its next-generation general humanoid foundation model, UnifoLM-WLA-1.0, making the code, model weights, and accompanying datasets publicly available simultaneously. The model's core training architecture is built on large-scale multimodal perception understanding and interaction-centric world modeling, optimized using approximately 2,500 hours of real-world robot operational data. In terms of performance, UnifoLM-WLA-1.0 leads among open-source models in various embodied intelligence reasoning benchmarks, with capabilities comparable to top-tier closed-source models in the industry. In practical applications, the model demonstrates strong generalization potential, capable of managing 64 distinct tasks through a single model. Its scenarios range from precise desktop manipulation to whole-body mobility control, supporting flexible adaptation across different task types and end-effectors.

## Sources

- [Unitree宇树](https://mp.weixin.qq.com/s/TkbZMqKTIS5FVqBdzO2a7A) (2026-09-15)

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.