Conceptual diagram of AI Photo: VCG

Qwen’s new model, Qwen3.6-Plus, topped the daily rankings on the widely recognized global large-model API platform OpenRouter on Saturday, according to the platform’s official website. 

The model just launched on Thursday. As of the press time, Qwen3.6-Plus recorded more than 1.4 trillion tokens in daily usage, data from OpenRouter showed, with 36Kr reporting that the figure set a global record for the highest single-day usage by a single model on OpenRouter.

Although industry insiders noted that the platform accounts for a limited share of global commercial API usage, it still highlighted the strong performance-to-cost appeal of these models.

As a key unit underpinning AI computation and interaction, token usage serves as an important indicator of a model’s real-world adoption, inference demand, and commercial traction. 

Notably, Chinese large language models have recently dominated a range of global rankings, with newly released models consistently demonstrating strong capabilities and standing out in an increasingly competitive landscape. 

On Friday, Code Arena, a programming-focused benchmark under the globally recognized LMArena platform, released its latest rankings. In the “Lab Ranking,” Alibaba, with its Qwen3.6-Plus model, ranked second globally, while the model itself placed sixth worldwide, the highest-ranking Chinese large language model, according to Code Arena’s official website.

The ranking is based on AI models’ performance in agentic coding tasks, assessing their capabilities in multi-step reasoning and tool use, according to Code Arena. Alibaba outperformed international tech giants including OpenAI, Google and xAI in the ranking, making it the highest-placed Chinese company on the list.

Among the top 10 companies in the rankings, five were from China, including Xiaomi, Moonshot, MiniMax and DeepSeek. 
 
Notably, among its standout models, China Media Group (CMG) reported on March 23 that Chinese large model MiniMax M2.5 had topped the global large-model usage rankings for five consecutive weeks.

Meanwhile, another Chinese AI startup GigaAI’s latest model, GigaWorld-1, climbed to the top of WorldArena, an authoritative evaluation platform for world models, outperforming models from Google and Nvidia.

It is also one of only two embodied world models to achieve an overall score above 60. On the March 30 leaderboard, it was the only model to break the 60-point mark.
 
Such achievements reflect the nation’s rapidly advancing pace in AI development and broader technological progress, industry observers noted. As China continues to push forward across the AI sector, beyond rankings, indicators such as token usage are increasingly highlighting both the depth of real-world deployment and the growing global competitiveness of its homegrown models.

As of March 15, Chinese AI large models recorded 4.69 trillion tokens in weekly usage, surpassing the US for a second consecutive week, according to OpenRouter data, as cited by CMG. The top three spots in global usage rankings were all held by Chinese models, the data showed.

In the domestic market, token demand has also skyrocketed, partly driven by China's "AI plus" initiative and the explosive growth of agent-based deployment, according to Xinhua News Agency. 

Official data showed that as of March this year, China’s average daily token calls had exceeded 140 trillion, up more than 1,000-fold from 100 billion in early 2024. Compared with 100 trillion at the end of 2025, the figure rose by more than 40 percent in just three months.

Chinese large models currently hold an edge in optimization and strong engineering execution. With computing resources constrained, developers have pursued a differentiated path from closed-source US models by relying on open-source approaches and low-level optimization to “deliver more with less,” Chen Jing, a vice president of the Technology and Strategy Research Institute, told the Global Times on Saturday.

In terms of the industrial chain, China still faces constraints in computing power. On the commercialization front, Chinese users are beginning to pay more actively, and the growth trend is encouraging, but there remains a gap with leading countries in the maturity of the commercial ecosystem, Chen said. 

To underpin AI growth, China's Government Work Report this year unveiled plans to launch new infrastructure projects focused on hyperscale intelligent computing clusters, alongside the coordinated development of computing capacity and power supply, according to Xinhua.

According to CMG, citing JPMorgan, China’s AI inference token consumption is projected to rise from about 10 quadrillion in 2025 to around 3,900 quadrillion by 2030, an approximately 370-fold increase over five years.