AI Photo: VCG
China’s open-source artificial intelligence (AI) models, such as Alibaba Cloud’s Qwen and DeepSeek, are rapidly narrowing the performance gap with top global proprietary models, as open-source models become a major growth engine in the enterprise large-language model market, the Xinhua News Agency said on Monday, citing a market survey.
The latest report from market research firm Frost & Sullivan, which conducted a comprehensive study of enterprise-level large-language model (LLM) usage in China’s generative AI market for 2025, indicates that the enterprise LLM sector is experiencing explosive growth.
In the first half of 2025, the daily total token consumption of enterprise LLMs in China reached 10.2 trillion, up 363 percent compared with the second half of 2024. Alibaba’s Tongyi led the market with a 17.7 percent share, making it the most widely adopted LLM among Chinese enterprises, according to Xinhua.
The progress comes as generative AI rapidly penetrates the Chinese enterprise market. The report showed that 70 percent of enterprises opted for public cloud deployment or access for LLMs, and 71 percent expressed plans to increase their use of public cloud-based generative AI services.
These findings further show that Chinese companies are shifting from pursuing the single most powerful LLM to seeking the optimal solution for specific business scenarios.
Open-source models have emerged as a critical driver of new growth in the enterprise LLM market. With continued open-source releases of domestic models like Qwen and DeepSeek throughout 2025, the performance gap between open-source models and leading international proprietary models has nearly vanished, Xinhua said, citing the report.
The Frost & Sullivan report predicted that more than 80 percent of enterprises will adopt open-source LLMs, indicating that open-source solutions will dominate growth in industry applications.
“China’s AI technology has advanced rapidly in recent years. In the large model domain, the country not only offers a wide range of models but has also achieved technological leadership, driven in part by its open-source approach, which has created a cluster advantage,” Ma Jihua, a veteran telecom industry analyst, told the Global Times on Monday.
Recently, Alibaba’s Tongyi open-sourced several new models, including the instruction-tuned Qwen3, inference-optimised models, as well as multimodal models such as the AI programming model Qwen3-Coder, according to the company.
Alibaba’s Qwen model has captured more than 12.3 percent of the global market, surpassing US-based models like OpenAI and the Llama series to rank fourth worldwide, according to data from OpenRouter, an international API aggregation platform for large models, as reported by Xinhua.
Ma emphasized that China’s complete industrial chain and strong manufacturing capabilities enable companies to adopt open-source strategies to build an AI ecosystem.
“The open-source approach draws developers worldwide, speeding up technological upgrades and real-world deployment, and it could enable open-source models to overtake proprietary ones in reach and use cases,” Ma said, adding that by lowering adoption barriers, open source also helps democratize AI and drive its global progress.
China’s open-source push is gaining momentum, with more domestic tech firms making their proprietary technologies publicly available. In early August, tech giant Huawei fully open-sourced its Ascend AI hardware framework, CANN, along with the Mind series of application suites and development tools.
The rapid rise of China’s open-source LLMs signifies a profound shift in the global AI landscape, as they boast advantages in scale, performance, ecosystem development, and commercial application deployment, Ma noted.
According to the national next-generation AI development plan, China’s core AI industry is expected to exceed 1 trillion yuan ($140.9 billion) in scale, with related industries surpassing 10 trillion yuan, according to Xinhua.