China’s open-source artificial intelligence models accounted for nearly 30 per cent of total global use of the technology, while Chinese-language prompts ranked second in token volume behind English, according to a report.
This year’s surge in open-source large language model (LLM) usage around the world had been fuelled by Chinese-developed systems, including Alibaba Group Holding’s Qwen family of models, DeepSeek’s V3 and Moonshot AI’s Kimi K2, according to a recently published report by OpenRouter, a third-party AI model aggregator, and venture capital firm Andreessen Horowitz. Alibaba owns the South China Morning Post.
Proprietary Western models, such as OpenAI’s GPT-4o and GPT-5, remained dominant with a 70 per cent global share.
Advertisement
According to the empirical study of 100 trillion tokens by OpenRouter, Chinese open-source LLMs’ global share started from a low base of 1.2 per cent in late 2024 to reach nearly 30 per cent over a few months this year. Tokens are units of data processed by AI models during training and inference, enabling prediction, generation and reasoning.
So far this year, Chinese open-source LLMs averaged 13 per cent of weekly token volume, as growth accelerated in the second half of 2025, to almost match the 13.7 per cent average recorded by AI models from the rest of the world, the report said.
Advertisement
“China has emerged as a major force, not only through domestic consumption but also by producing globally competitive models,” the report said.
The report offered fresh evidence that China has become a close peer of the United States in AI model development, despite Washington’s restrictions on Chinese firms’ access to advanced graphics processing units from the likes of Nvidia and Advanced Micro Devices.