---
slug: "sota-f4dd7961"
title: "Ascend computing power has trained an embodied world model that achieves SOTA on two leaderboards, and Huawei Cloud now wants to open up the entire foundation stack."
headline: "Ascend computing power has trained an embodied world model that achieves SOTA on two leaderboards, and Huawei Cloud now wants to open up the entire foundation stack."
canonical: "https://silinex.xyz/wire/sota-f4dd7961"
date_published: "2026-11-20T08:04:58.000+00:00"
date_modified: "2026-11-20T10:20:29.436+00:00"
author: "量子位"
publisher: "Silicon Nexus"
category: "robotics"
summary: "Compiled from public reports including Quantum位: Huawei Cloud is about to open-source its embodied world model—and it's launching straight to first place on two leaderboards!"
signal_type: "research"
impact_score: 5
source_tier: "L3"
tldr:
  - "Compiled from public reports including QuantumBit: Huawei Cloud is open-sourcing its embodied world model—and it debuts at the top of two leaderboards at once!"
  - "On the timeline: on November 16, Huawei Cloud released the Ascend-native embodied world model CloudRobo-WM1.2."
  - "In two embodied intelligence benchmarks, RoboTwin 2.0 and LIBERO-Pro 2.0, it ranks first in average success rate on both."
entities:
  []
key_metrics:
  []
key_facts:
  - "Compiled from public reports including QuantumBit: Huawei Cloud is open-sourcing its embodied world model—and it debuts at the top of two leaderboards at once!"
  - "On the timeline: on November 16, Huawei Cloud released the Ascend-native embodied world model CloudRobo-WM1.2."
  - "In two embodied intelligence benchmarks, RoboTwin 2.0 and LIBERO-Pro 2.0, it ranks first in average success rate on both."
brief_body: "Compiled from public reports including Quantum位: Huawei Cloud is about to open-source its embodied world model—and it's launching straight to first place on two leaderboards!\n\nTimeline: On November 16, Huawei Cloud released the Ascend-native embodied world model CloudRobo-WM1.2. It ranked first in average success rate on two embodied intelligence benchmarks, RoboTwin 2.0 and LIBERO-Pro 2.0. Huawei Cloud plans to open the model weights, inference code, and post-training code on the same day.\n\nPublic information also notes: Well, open-sourcing even the post-training code outright—Huawei's move is indeed aggressive. Other robotics teams that want to teach this model a new task no longer need to build a training pipeline from scratch; with the ready-made code, they can plug in their own data and start debugging and validating right away! And this time, Huawei Cloud did not stop at \"getting the model to run on Ascend chips.\" Instead, from model architecture and training framework to core operators and inference deployment, it optimized software-hardware co-design around the characteristics of embodied models. From world model pre-training and reinforcement learning to robots getting hands-on with tasks, Huawei Cloud has connected the entire pipeline."
---
# Ascend computing power has trained an embodied world model that achieves SOTA on two leaderboards, and Huawei Cloud now wants to open up the entire foundation stack.

## Silicon TL;DR

1. Compiled from public reports including QuantumBit: Huawei Cloud is open-sourcing its embodied world model—and it debuts at the top of two leaderboards at once!
2. On the timeline: on November 16, Huawei Cloud released the Ascend-native embodied world model CloudRobo-WM1.2.
3. In two embodied intelligence benchmarks, RoboTwin 2.0 and LIBERO-Pro 2.0, it ranks first in average success rate on both.

## Wire brief

Compiled from public reports including Quantum位: Huawei Cloud is about to open-source its embodied world model—and it's launching straight to first place on two leaderboards!

Timeline: On November 16, Huawei Cloud released the Ascend-native embodied world model CloudRobo-WM1.2. It ranked first in average success rate on two embodied intelligence benchmarks, RoboTwin 2.0 and LIBERO-Pro 2.0. Huawei Cloud plans to open the model weights, inference code, and post-training code on the same day.

Public information also notes: Well, open-sourcing even the post-training code outright—Huawei's move is indeed aggressive. Other robotics teams that want to teach this model a new task no longer need to build a training pipeline from scratch; with the ready-made code, they can plug in their own data and start debugging and validating right away! And this time, Huawei Cloud did not stop at "getting the model to run on Ascend chips." Instead, from model architecture and training framework to core operators and inference deployment, it optimized software-hardware co-design around the characteristics of embodied models. From world model pre-training and reinforcement learning to robots getting hands-on with tasks, Huawei Cloud has connected the entire pipeline.

---

Source desk: **量子位** · [origin link](https://www.qbitai.com/2026/11/511610.html)
