Huawei Defends Pro MoE LLM Model Amid Accusations of Deriving from Alibaba’s Qwen-2.5

Huawei Technologies, one of China’s leading technology giants, has strongly denied recent allegations that its newly launched open-source large language model (LLM), known as Pro MoE, was developed through incremental training or fine-tuning on existing third-party AI models—specifically Alibaba’s Qwen-2.5 model. Instead, Huawei maintains that the Pro MoE model is an original creation, developed entirely using its own infrastructure, algorithms, and proprietary Ascend AI chip technology.

The controversy surfaced when a well-known open-source AI community member, using the handle HonestAGI, published claims on GitHub suggesting that Huawei’s Pangu Pro MoE 72B model showed an “extraordinary correlation” with Alibaba’s Qwen-2.5 14B model. This assertion immediately sparked a wave of discussion and skepticism within the AI research and development community, raising serious questions about the originality of Huawei’s model and its training methods.

In response, Huawei’s Noah’s Ark Lab—the division responsible for the development of the Pangu series of AI models—issued a clear and detailed statement on Saturday. The lab emphasized that the Pangu Pro MoE model was developed and trained entirely on Huawei’s Ascend AI chip platform, a hardware ecosystem designed and manufactured in-house. They categorically denied that the model was the product of incremental or transfer learning using any other company’s AI models, stating:

“The Pangu Pro MoE model is a result of original development efforts on Huawei’s Ascend hardware platform. It is not derived from or built upon any third-party AI models through incremental training.”

Clarifying Use of Open-Source Components

Huawei also addressed the issue of code usage, acknowledging that certain portions of their model’s development utilized open-source code, which is common practice in the AI research community. However, the company stressed that all such code usage complied fully with applicable open-source licenses. Furthermore, Huawei confirmed that any external code was properly labelled and credited in accordance with legal requirements, dismissing accusations that they had misappropriated proprietary work.

The initial GitHub repository posting the allegations has since been removed, leaving behind only a brief explanation without technical evidence to substantiate the claimed similarity between the two models. To date, no conclusive technical reports or comparative model evaluations have been released publicly to verify or disprove these claims.

Rising Competition in China’s LLM Market

Huawei’s firm defense of its Pro MoE model occurs amid intensifying competition in the rapidly evolving Chinese AI landscape. Several key players, including Alibaba, DeepSeek, and other startups, have been aggressively advancing their own large language model initiatives, striving to establish dominance in the field of generative AI.

Alibaba’s Qwen-2.5 model is widely regarded as one of the more advanced open-source LLMs developed in China, making it a natural benchmark for comparison. The proximity of development timelines, shared open-source AI tools, and industry trends can sometimes lead to perceived similarities across models, which further fuels debate among researchers.

Huawei, frequently spotlighted as a symbol of China’s resilience against U.S. tech sanctions and restricted access to foreign semiconductor technology, has heavily invested in developing indigenous AI capabilities. Its Ascend chip family plays a crucial role in enabling the company’s AI ambitions, from foundational research to commercial applications.

The Importance of Transparency and Trust

In the fast-moving world of AI, where breakthroughs and product launches often race ahead of formal peer review, questions about originality and ethical development have become paramount. Huawei’s proactive public statement serves to reassure its partners, clients, and the broader AI community of its commitment to transparent and responsible AI development practices.

The Noah’s Ark Lab’s reaffirmation that the Pro MoE model is a standalone product signals Huawei’s desire to maintain credibility and lead innovation in the sector, despite the scrutiny. This move also highlights the growing emphasis placed by governments, developers, and enterprises on open-source compliance, algorithmic integrity, and trustworthiness in AI.

Looking Forward: The Future of Chinese AI Models

As AI research continues to accelerate globally, and open-source models become increasingly central to innovation, incidents like this highlight the competitive dynamics and collaboration tensions inherent in the field. Chinese companies are rapidly closing the gap with Western AI leaders, propelled by massive investment, government support, and a robust talent pipeline.

Huawei’s advancements with the Pro MoE model and its Ascend AI chip ecosystem demonstrate the company’s strategy to build end-to-end AI solutions within China’s evolving technology ecosystem. Whether the allegations of model similarity will spur deeper technical audits or foster greater transparency remains to be seen.

For developers, enterprises, and governments watching the rise of AI capabilities, the episode underscores the critical importance of due diligence, independent verification, and clear communication when releasing influential AI technologies in the public domain.