A photo shows the logos of Alibaba and its AI model application Qwen Photo: VCG

Chinese AI models are becoming indispensable tools for an increasing number of overseas developers, the country's industry observers said on Sunday. Despite US' export controls and politically motivated restrictions, Alibaba's Qwen, DeepSeek, Moonshot AI's Kimi, Zhipu AI and MiniMax are expanding rapidly in the world, according to experts.

Alibaba Group's open-source AI model Qwen has reportedly surpassed 3 billion downloads worldwide during the past six months, overtaking Meta and Google to claim the top spot globally, according to an August 14 report from open-source platform Hugging Face.

The Hugging Face report showed that Google's open models had 418 million downloads, and Meta's had 227 million on the platform in the same period.
 

These figures highlight the accelerating adoption of Chinese open-source and open-weight models, despite US' export control measures, Tian Feng, former dean of SenseTime's Intelligence Industry Research Institute, told the Global Times on Sunday. 

According to the Qwen team, Alibaba has open-sourced more than 460 models through the Qwen family, and the derivative models built on top of them now exceed 300,000.

In its report, Hugging Face described Qwen as "one of the largest foundations of the open AI ecosystem," noting that "Qwen has become embedded in the default workflow of developers deciding which model to fine-tune and deploy."

Open-weight models can be downloaded, fine-tuned and used as the foundation for building new AI products. Their adoption has become an important indicator of which technologies global developers prefer when selecting a base model.

According to the Hugging Face report, in almost every month of 2026, the largest and most performant open model released by a Chinese lab was larger than anything released by an American lab. China's monthly ceiling ranged from 754 billion to 2.78 trillion parameters, while America's ceiling stayed below 130 billion in five of the seven months covered by the report.

Industry observers said Qwen's rise to the top suggests that the strategy of offering relatively inexpensive, easily adaptable models is being accepted by developers and users, which signals that China's AI industry has moved beyond merely catching up and entered a phase in which it is setting standards for the global open-source ecosystem.

Qwen, along with Moonshot AI, DeepSeek and other Chinese AI models, is replicating frontier-level performance, seeking to bridge the gap with closed models in US, such as those from OpenAI and Anthropic. US export controls on chips and AI systems, such as the brief ban on overseas access to Anthropic's Fable 5 model this summer, don't appear to be putting the brakes on Chinese competitors, according to Bloomberg.

Chinese AI companies have released major open models recently. On Friday, Zhipu AI launched GLM-5.3, its latest foundation model, saying the open-source release would make security defense capabilities a public resource that global developers can improve together, the Science and Technology Daily reported.

On Thursday, DeepSeek released Harness, its first agent runtime framework. The framework allows models to directly interact with code repositories, locate files, modify code, run tests and iteratively fix errors.

Moonshot AI recently announced the full open-sourcing of Kimi K3, its latest flagship model. With 2.8 trillion total parameters, it is currently the world's largest open model by parameter scale.

As competition in the developer market intensifies, US tech companies Meta and Nvidia have released new open-source AI models in recent weeks for developers to download through the open-source ecosystem, in contrast to the proprietary models offered by OpenAI and Anthropic, according to CNBC.

Open weights are expected to strengthen competition among model developers and across cloud computing, chips, applications and services, Tian said. Such competition will inspire innovation, reduce costs and make advanced AI systems more accessible, adaptable and widely available.

Major US technology companies and institutions, including Nvidia, Microsoft, IBM, Meta, OpenAI, Cisco, Dell Technologies, GitHub, and Hugging Face have recently issued a joint statement, calling on US policymakers to support open-weight AI models and avoid restrictions. They argued that US leadership in AI depends on building an open ecosystem across all sectors.

Recently, a group of nearly 200 Silicon Valley-based startups, calling themselves the Little Tech Association, jointly sent a letter to the US government, opposing restrictions on American companies' use of Chinese open-weight AI models.

These moves indicate a vote of confidence of global developers and users in China's open-source AI ecosystem, Liu Gang, chief economist at the Chinese Institute of New Generation Artificial Intelligence Development Strategies, told the Global Times on Sunday.

In fact, Chinese open models are being widely used in the world and are helping narrow the digital divide. Their use in agriculture, education and healthcare has brought tangible benefits to the developing countries across the Global South, Liu said.

Taking Alibaba's Qwen for an example, the company has been distributing models to enterprise customers in Southeast Asia and Africa through its cloud platform, further strengthening its global footprint. The Qwen family covers both text and multimodal tasks and supports 119 languages and regional dialects, according to the Qwen team.

According to Liu, the widespread use of Chinese models shows that they have already become part of the global AI supply chain. "In the global wave of AI development, China's open-source models are infusing it with more of the genes of openness, sharing and mutual benefit," Liu said.

Global AI model usage reached 69 trillion tokens in the week from August 3-9, up 21.48 percent week-on-week, according to calculations by National Business Daily based on OpenRouter's latest data. Chinese models accounted for 34.25 trillion tokens in weekly usage, rising 21.76 percent from the previous week and surpassing US models for the 15th consecutive week to rank first globally.