Huawei denies reports that its open source AI model 'Pangu Pro MoE' is 'recycled from Alibaba'



The large-scale language model ' Pangu Pro MoE 72B ' with 72 billion parameters, released on June 30, 2025, was accused of being a reused product from Alibaba, and the developer, Huawei, publicly denied this.

华为盘古团队未认AI model 抄袭阿通义千问 |
https://www.zaobao.com.sg/realtime/china/story20250707-7092327

Huawei pushes back on AI model plagiarism claims
https://the-decoder.com/huawei-pushes-back-on-ai-model-plagiarism-claims/

Huawei's AI Lab Fends Off Accusations It Copied Rival Models - Bloomberg
https://www.bloomberg.com/news/articles/2025-07-07/huawei-s-ai-lab-fends-off-accusations-it-copied-rival-models

The Pangu Pro MoE 72B is a model announced by Noah's Ark Lab, Huawei's AI research division. It has demonstrated high performance using the 'Ascend' series of chips built without relying on the United States, and was praised for announcing its own high-performance model despite being subject to US regulations.

Huawei releases 'Pangu Pro MoE 72B', a language model trained in China's AI ecosystem, and open-sources inference technology - GIGAZINE



However, suspicions of fraud arose about this model.

On July 3, 2025, an unidentified user named HonestAGI posted an English paper on GitHub, pointing out that Pangu Pro MoE 72B showed an extraordinary correlation with Alibaba's Qwen 2.5 14B . The paper even stated that 'Noah's Ark Lab did not train the model from scratch, but instead reused Qwen 2.5 14B,' sparking widespread discussion among AI users. The paper was later deleted.




In addition, another user claiming to be an employee of Noah's Ark Lab published a statement on GitHub, similarly claiming that 'Noah's Ark Lab trained the model using Alibaba's model.'

GitHub - HW-whistleblower/True-Story-of-Pangu: A genuine dark story behind the ancient large model study.
https://github.com/HW-whistleblower/True-Story-of-Pangu

In response to this uproar, Noah's Ark Lab issued a statement completely denying the allegations. Noah's Ark Lab stated, 'Pangu Pro MoE is a large-scale model developed and trained based on Ascend, and is not based on models from other vendors. Some code has been quoted in accordance with industry open source practices, but all licenses are properly indicated.'

Regarding the Chinese AI, there are suspicions that 'DeepSeek-R1' was trained using models from other companies, but these are merely speculations.

Speculations are rife that the Chinese-made high-performance AI model 'DeepSeek-R1-0528' may have been distilled using Google's AI 'Gemini' - GIGAZINE



in Software, Posted by log1p_kr