Chinese open-source AI models are winning over developers worldwide
From DeepSeek to Alibaba’s Qwen, openly released Chinese AI models now lead global download charts — opening low-cost opportunities for developers across the Global South.
Chinese artificial intelligence labs have quietly become the world’s leading suppliers of open AI models. According to reporting by MIT Technology Review, Chinese open-source models have now surpassed US models in total downloads, a shift that began with the release of DeepSeek’s R1 reasoning model in January 2025.
DeepSeek released R1 as an open-weight model under a permissive MIT licence and published its training methods. The model performed comparably to leading Western systems at a fraction of the cost, briefly topped the US App Store and triggered a sell-off that temporarily wiped around $1 trillion off US technology stocks.
Qwen leads the downloads
Alibaba’s Qwen family — ranging from laptop-sized models to large data-centre systems — was the most downloaded model series on Hugging Face in 2025 and has overtaken Meta’s Llama in cumulative downloads. By August 2025, Qwen-based derivatives made up more than 40% of new language-model derivatives on the platform, compared with roughly 15% for Llama.
Newer releases continue the trend. Moonshot AI’s Kimi K2.5, launched in early February 2026, came close to top proprietary models on some early benchmarks at around one-seventh of the price. Two other Chinese labs, Z.ai and MiniMax, listed on the Hong Kong stock exchange in January 2026.
Why it matters for the Global South
For developers and start-ups in countries such as Pakistan, open models offer a practical route into AI: they can be downloaded, run on local servers and adapted for local languages and needs without paying high per-use fees to overseas providers. Chinese labs are also moving towards smaller, specialised models that run on ordinary devices, further lowering costs.
The caveats
The ecosystem is not fully independent. Many Chinese models still rely on Nvidia chips and US cloud platforms, and the long-term sustainability of free releases depends on the companies’ finances. Even so, the open-model race has given the rest of the world more choice — and more bargaining power — than it had two years ago.
Sources: MIT Technology Review
