Artificial intelligence Photo: VCG

Chinese large language models, shorten as LLMs, are once again generating significant buzz and making headlines in the global AI community and international media. 

In the past week, three Chinese LLMs continued to rank among the top five in global token usage on the OpenRouter leaderboard, according to a report by the Xinhua News Agency. OpenRouter is one of the major overseas multi-model aggregation and routing platforms, serving as an important window for observing global large model usage trends and overall popularity.

A token is the smallest unit of text that an AI language model reads and processes. It can be a whole word, part of a word, a punctuation mark, or even a single character. 

AI models by Chinese companies have overtaken the US in weekly token consumption for the third consecutive week from March 16 to March 22, the Beijing-based Science and Technology Daily reported on Tuesday. 

Among the top 10 AI models on the leaderboard, Chinese AI models recorded a weekly token usage of 7.359 trillion tokens, representing a 56.9 percent increase compared to the previous week. 

On March 16, Elon Musk commented "impressive work from Kimi" on X, highlighting a major breakthrough by Moonshot AI, Kimi's developer, on "attention residuals," a novel improvement to the Transformer architecture that enhances training efficiency and model performance at scale.

The Chinese AI company's overseas revenue reportedly has surpassed its domestic revenue since it released K2.5 AI LLMs in late January, signaling that its commercialization efforts have begun to compete on the global stage, analysts said. 

Kimi's recent global commercialization offers a glimpse into how an increasing number of Chinese AI LLMs have seen their global influence rise in recent months. 

In addition to Kimi, Chinese large models including MiniMax, DeepSeek and Zhipu, as well as Doubao, Hunyuan and Qwen - respectively backed by leading Chinese tech companies ByteDance, Tencent and Alibaba - have begun what some industry observers describe as "an epic expedition" in the global market.

'An inevitable momentum'

From a task-structure perspective, the way the Chinese LLMs are being used overseas is also undergoing a significant shift. 

According to OpenRouter's report, programming and technical tasks now account for a combined 39 percent of usage among Chinese open-source large models. This indicates that domestic models are increasingly moving into code generation, technical support, and infrastructure-related scenarios, rather than remaining primarily focused on lighter tasks such as creative generation, the Xinhua report said.

"In the past, our models were mainly used by programmers, but now people from all walks of life around the world are eagerly experimenting with them and integrating our AI agents into their work and daily lives," MiniMax's vice president Yan Yijun told the Global Times on Tuesday. 

According to Yan, the MiniMax M2.5 model has consistently topped the OpenRouter usage rankings since February, showing that Chinese LLMs have been "expanding on the international stage on a massive scale."

According to MiniMax's 2025 full-year performance report, over 70 percent of the Chinese AI start-up's revenue in 2025 was generated from the international market. The company's native products and open platform have cumulatively served over 236 million individual users across more than 200 countries and regions, as well as 214,000 enterprise clients and developers from over 100 countries and regions.

"It is just a prelude," said Tian Feng, former dean of SenseTime's Intelligence Industry Research Institute, referring to the streak of Chinese LLMs' overseas expansion. 

Tian told the Global Times on Tuesday that the trend "is inevitable as Chinese LLMs demonstrate strong competitiveness in delivering top-tier coding and logical reasoning capabilities, while also bringing down the input and output costs to a highly disruptive level."

According to Artificial Analysis' latest evaluation of mainstream LLMs, several Chinese models have entered the global top tier in intelligence performance. The gap with leading international models continues to narrow in areas such as programming ability, multimodality, and long-context reasoning, the Xinhua report said.

Yan said that the Chinese LLMs, while matching the capabilities of leading international models, offer prices that are just one-tenth or even less of theirs. 

"When overseas users discover that our models not only genuinely improve their productivity but are also priced at a truly affordable, inclusive level, they naturally 'vote with their feet,'" Yan said.

Analysts also stressed that the rise of agent tools represented by OpenClaw has accelerated the adoption of LLMs in high-frequency scenarios such as programming and office work, which requires longer task chains, more continuous calls, and greater memory requirements, leading to a significant increase in token consumption. Those developments further highlight the "unparalleled cost-performance advantage" of the Chinese LLMs. 

Commercialization underway

At the beginning of 2024, China's daily token usage was around 100 billion, rising to 100 trillion by the end of 2025, and surpassing 140 trillion in March this year - an increase of over a thousandfold in two years, the People's Daily reported on Tuesday.

Liu Liehong, head of the National Data Administration, noted that since the end of January this year, some model companies have generated in just 20 days more revenue than their total for all of 2025, highlighting the rapid development of a new token-based business model, according to the report. 

"When the LLM technology emerged two years ago, what we often heard in our discussions with AI start-ups were technical terms, such as 'our model is comparable to ChatGPT in parameter scale,' and 'our multimodal capabilities make a stunning debut,'" a private equity fund manager surnamed Zhu based in Shenzhen, South China's Guangdong Province, told the Global Times on Tuesday.

"But starting this year, what we [as investors] are looking for are early signs of commercialization: We often ask AI start-ups questions such as 'how many paying customers do you have?' 'what is your API call volume?'" Zhu said. 

Industry observers believed that overseas expansion offers one of the options for AI start-ups to speed up the commercialization process. 

"In the domestic market, competition in the Chinese AI ecosystem has been growing white-hot, which sharply limits profit margins," Tian said. "Within the AI start-ups themselves, while revenue is growing rapidly, the cost of computing infrastructure is also surging rapidly." 

Tian meanwhile stressed that in overseas markets such as the US and Europe, per capita purchasing power could be significantly stronger compared with the domestic market.

Yan believed that the globalization of "homegrown tokens" will definitely see explosive growth in the future, which then could "profoundly reshape the ecosystem of global LLMs."

"In the past, the massive AI demand from Europe, Southeast Asia, and the Middle East could only be met by a handful of US companies. But now, Chinese AI models have entered the arena, demonstrating exceptional cross-lingual and cross-cultural capabilities," Yan noted, "This has not only broken the previous technological monopoly, but also signals that the global AI landscape is gradually entering an era of parallel leadership between China and the US."