Chinese-developed artificial intelligence models now dominate usage on the world’s largest LLM API aggregation platform. OpenRouter data published on February 24, 2026, shows that models built in China account for 61% of total token consumption. These models are primarily driving usage in programming and agent-driven workflows.
Weekly statistics confirm that the top three most-used models on the platform were developed by Chinese AI laboratories. The aggregate consumption of Chinese models reached 5.3 trillion tokens, contributing to a total of 8.7 trillion tokens consumed by the top 10 models. This indicates a significant concentration of usage among domestic Chinese offerings compared to international competitors.
MiniMax M2.5 secured the first position with 2.45 trillion tokens consumed in a single week. According to a report by China’s Science and Technology Innovation Board Daily, this figure represents a 197% increase compared to the prior week. Moonshot AI’s Kimi K2.5 followed in second place with 1.21 trillion tokens, although its usage declined 20% week-over-week. Zhipu AI’s GLM-5 took third place with 780 billion tokens, marking a 158% surge. DeepSeek V3.2 ranked fifth overall.
The rapid ascent of MiniMax M2.5 was partially fueled by promotional access. Kilo Code offered the model free of charge for a week starting February 12. Cline, another developer tool, ran a similar promotion. Despite the impact of these giveaways on raw token volume, the model demonstrates competitive performance. MiniMax M2.5 scores 80.2% on SWE-Bench Verified, a benchmark for software engineering tasks. This result is close to the 80.8% score achieved by Anthropic’s Claude Opus 4.6.
A structural shift in application focus underpins the consumption data. Programming has evolved into the largest category of token usage on OpenRouter, expanding from 11% to over 50% of the total throughout 2025. Furthermore, agent-driven workflows, which involve models autonomously executing multi-step tasks, now generate more than half of all output tokens on the platform.
OpenRouter COO Chris Clark noted the specific role of Chinese models in this trend. He stated that Chinese open-weight models have captured significant market share because they are “disproportionately heavy in agentic flows run by U.S. firms.” All three top-ranked Chinese models this week are specifically optimized for coding and agent automation, aligning with these usage patterns.
Pricing disparities provide a strong incentive for adoption. MiniMax M2.5 costs $0.30 per million input tokens and $1.10 per million output tokens. Zhipu AI’s GLM-5 is priced at $0.30 and $2.55 respectively. In contrast, Claude Opus 4.6 costs $5 per million input tokens and $25 per million output tokens, making it roughly 10 to 20 times more expensive than its Chinese counterparts.
OpenRouter’s overall scale reflects the rapid expansion of the sector. Total weekly token consumption on the platform has reached approximately 12.1 trillion, a 12.7-fold increase from the previous year. Founded by former OpenSea CTO Alex Atallah, the platform provides a unified API gateway to more than 400 models from over 60 providers. As MIT Technology Review reported, this adoption is extending to Silicon Valley startups. Andreessen Horowitz partner Martin Casado estimated that roughly 80% of startups utilizing open-source AI stacks are running Chinese models.
