Huawei Denies Model AI Pangu Plagiarizing Alibaba's Qwen

Huawei Pangu Pro Moe is accused of plagiarizing Alibaba's model, Qwen 2.5-14B. (photo: x @China__Focus)

JAKARTA Huawei's artificial intelligence research division has denied allegations that one of its large language models (LLM), namely Pangu Pro Moe (Mixture of Experts), has plagiarized Alibaba's model, Qwen 2.5-14B. Huawei stated that the model was developed and trained independently.

This statement was released by Noah Ark Lab, Huawei's AI division, on Sunday 6 July, a day after an entity named HonestAGI uploaded a paper in English on the code-sharing platform Github. In the paper it is stated that the Pangu Pro Moe model exhibits extraordinary correlations' with Alibaba's 2.5-14B Qwen model.

The paper states that the significant resemblance indicates that Huawei's model was not trained from the start, but is the result of an "recycle", thus triggering widespread discussions in the global AI community and technology media in China. HonestAGI also alleges potential copyright infringement, falsification of technical information, as well as false claims related to Huawei's investment in training the model.

In response to this, Noah Ark Lab confirmed that the Pangu model was not built from additional training on models belonging to other parties, and they had made important innovations in architectural design and its technical features. The model is also referred to as the first large-scale LLM to be fully built on the Huawei Ascend chip.

In its official statement, Noah Ark Lab also added that its development team has strictly complied with open-source licenses in the use of third-party codes, although it does not explain which open-source models are referenced.

Alibaba has not yet responded to a request for comment from the media. HonestAGI has not been able to be contacted to find out who is behind the entity.

For information, Alibaba's model Qwen 2.5-14B was released in May 2024 as part of a family of small Qwen 2.5 models that can be run on PCs and smartphones. Meanwhile, Huawei, which has entered the LLM arena since 2021 with the release of its first Pangu, has recently been considered lagging behind its competitors.

At the end of June 2025, Huawei released the open-source Pangu Pro Moe model through China's developer GitCode, as part of its efforts to increase its AI technology adoption by providing free access for developers.

Unlike Qwen, which is more consumer-oriented, such as ChatGPT chatbot services, Huawei's Pangu models are mostly used in the government, finance, and manufacturing sectors.