DeepSeek and Qwen: how China turned open models into a leap forward in the global AI race

Chinese open source in AI has moved out of catch-up mode and become a full-fledged factor in global competition. Open models of Chinese origin are no longer just meeting domestic demand, but starting to influence which stacks, models, and products developers choose far beyond the country's borders.

Support from Above

In the early 2020s, open source became de facto the norm for Chinese corporate sector. According to CAICT research, more than 87% of companies used open tools, but only 24% of organizations allocated separate teams to manage open source. This meant a simple gap: technologies were already being applied widely, but processes, licenses, security, and internal expertise often lagged behind the scale of implementation.

For Beijing, this became not only an engineering but a strategic problem. The answer was a state-driven assembly of an ecosystem around open projects. In 2020, the OpenAtom foundation appeared with participation from Huawei, Alibaba, Baidu, and Tencent, and one of the anchor projects became OpenHarmony — an Android alternative, which the country views as the foundation for technological independence.

In parallel, Gitee grew, China's alternative to GitHub: by 2020, the platform already had more than 10 million repositories and about 5 million developers. It lags far behind GitHub, but for the domestic market, this was enough to create its own foundation for development.

From Frameworks to Models

China didn't jump straight to the LLM hype, but went through infrastructure. Even before the DeepSeek wave, the country developed its own deep learning frameworks: PaddlePaddle from Baidu, X-DeepLearning from Alibaba, and MindSpore from Huawei. These provided the local market with an environment where large models could be trained and deployed quickly without complete dependence on the American stack. Because of this, Chinese teams approached the generative AI era already equipped with a ready engineering base and applied scenarios in industry, agriculture, cloud, and enterprise software.

A few numbers show well why this breakthrough was noticed around the world:

• PaddlePaddle is used by more than 23 million developers from 760 thousand companies.
• Training DeepSeek-V3, according to the authors' estimates, cost about $5.5 million — orders of magnitude less than Western budgets for comparable models.
• Research on downloads of 851 thousand models showed that Chinese open-source models received 17.1% of downloads versus 15.8% for American ones.
• By September 2024, downloads of Qwen models exceeded 600 million worldwide.

The next stage is specialization. Chinese teams are increasingly releasing not just universal chat models, but systems for specific classes of tasks. Examples include Intern-S1-Pro for chemistry, biology, and materials science, Fleming-R1 for diagnosing rare diseases, and Tencent models for music generation and recognition. This is an important shift: competition is no longer just about model size or benchmark quality, but about how quickly open-weight solutions turn into working tools for industries.

The West Has Already Connected

The main effect is that Chinese open source has stopped being an internal story. According to OpenRouter, usage of Chinese open models has grown by approximately 30% in the past two years from nearly zero, with particularly strong growth in the second half of 2025. DeepSeek burst into the top chatbot downloads in the US within a week of release, and some Chinese projects began gaining Western audiences so quickly that they were forced to limit new subscriptions due to computational capacity constraints. Andreessen Horowitz partner Martin Casado describes the situation like this:

"Startups presenting their open-source projects work on

Chinese open models in roughly 80% of cases."

The behavior of the teams themselves is also changing. If previously Chinese developers rarely entered the English-language public sphere, now they go to Reddit, Western platforms, and international communities to promote their models directly. In parallel, they quickly monetize successful technical trends: when the OpenClaw agent framework took off in early 2026, Moonshot AI and MiniMax almost immediately offered ready-made solutions with minimal customization. This is no longer the position of a catch-up market, but the behavior of an ecosystem that knows how to quickly productize research achievements.

What This Means

Open AI models have become for China not a side direction, but a working mechanism of technological advancement. If the trend continues, the global market will face fiercer competition not only between closed American platforms, but also between open ecosystems, where China already plays the role of one of the centers of power.