Chinese open-source AI models capture nearly 30% of global usage

Open-source artificial intelligence models from China accounted for nearly 30 percent of total global use of the technology. Meanwhile, Chinese-language prompts ranked second in token volume behind English. This was according to a recently published report by OpenRouter, a third-party AI model aggregator, and venture capital firm Andreessen Horowitz.

The global increase in open-source large language model (LLM) adoption this year has been significantly driven by systems developed in China. These include prominent models such as the Qwen family from Alibaba Group Holding, DeepSeek’s V3, and Moonshot AI’s Kimi K2.

Chat models, such as OpenAI’s GPT-4o and GPT-5, remained dominant with a 70 percent global share.

Chinese open-source LLMs’ global share began from a low base of 1.2 percent in late 2024. It reached nearly 30 percent over a few months this year. This is according to the empirical study of 100 trillion tokens by OpenRouter. Tokens are units of data processed by AI models during training and inference, allowing prediction, generation, and reasoning.

Chinese AI models accounted for an average of 13 percent of the weekly token volume this year. Growth accelerated in the latter half of 2025. It brought their average usage close to the 13.7 percent recorded by AI models from the rest of the world, according to the report.

“China has emerged as a major force, not only through domestic consumption but also by producing globally competitive models,” the report stated.

The report showed new evidence that China has become a close equivalent of the United States in AI model development. This was despite Washington’s restrictions on Chinese firms’ access to advanced Graphics Processing Units (GPU) from the likes of Nvidia and Advanced Micro Devices (AMD).

The ascent of Chinese open-source AI models “reflects not only competitive quality but also rapid iteration and dense release cycles,” according to the report.

The rapid release schedules for Alibaba Cloud’s Qwen and DeepSeek’s models have allowed users to quickly adapt to higher development workloads. Alibaba Cloud is the AI and cloud computing services division of Alibaba, based in Hangzhou.

As China’s open-source models gain recognition for their increased efficiency and cost-effectiveness, Chinese has become the second most used prompt language worldwide. It accounts for nearly 5 percent of all requests, following the leading language, English.

That percentage was significantly higher than the Chinese language’s share on the internet. This stood at about 1.1 percent, according to various estimates.

China ranked fourth behind the US, Singapore, and Germany in terms of global share in LLM tokens, according to the report.

The global market for open-source AI models grew from a DeepSeek-led monopoly in December 2024. By late 2025, it had become a fragmented landscape. There was competition from Alibaba Cloud’s Qwen and Moonshot AI’s Kimi, according to the report. It mentioned LLM usage was now widely spread, with no single model surpassing a 25 percent share.