Chinese Models Lead Global LLM Usage
Chinese AI developers have surged ahead in the global language model race, capturing 61% of total token consumption on OpenRouter — the world’s largest large language model (LLM) aggregation platform. According to data released on February 24, 2026, the three most-used models on OpenRouter all originate from Chinese AI labs, underscoring their growing dominance in global AI workloads.
MiniMax M2.5 Breaks Records
At the top of the leaderboard, MiniMax M2.5 posted a record-breaking 2.45 trillion tokens in weekly usage — a 197% increase over the previous week. Data from China’s Science and Technology Innovation Board Daily shows Moonshot AI’s Kimi K2.5 in second place with 1.21 trillion tokens (down 20% week over week), followed by Zhipu AI’s GLM-5 with 780 billion tokens after a 158% surge. DeepSeek V3.2 rounded out the Chinese contingent in fifth position.
Collectively, Chinese models accounted for 5.3 trillion of the 8.7 trillion tokens consumed by OpenRouter’s top ten models. The spike in activity for MiniMax M2.5 was partly driven by temporary free promotions: AI coding platforms like Kilo Code and Cline offered complimentary access to the model in mid-February. Despite the promotional effect, interest in MiniMax M2.5 remains high — the model achieved an 80.2% score on the SWE-Bench Verified benchmark, nearly matching Anthropic’s Claude Opus 4.6 at 80.8%.
Coding and Agent Workflows Drive the Shift
The sharp rise of Chinese models reflects a broader trend in coding and agentic AI usage. Programming tasks now account for more than half of all token consumption on OpenRouter — up from just 11% in early 2025. At the same time, agent workflows, where models autonomously execute multi-step instructions, now generate the majority of output tokens.
OpenRouter COO Chris Clark told SaaStr that Chinese open-weight models are “disproportionately heavy in agentic flows run by U.S. firms.” The three leading Chinese models — MiniMax M2.5, Kimi K2.5, and GLM-5 — are all optimized for coding tasks and automation-driven applications, helping explain their rapid adoption.
Price Advantage Boosts Adoption
Pricing has played an equally critical role in this shift. MiniMax M2.5 costs $0.30 per million input tokens and $1.10 per million output tokens, while GLM-5 lists similar rates. In comparison, Anthropic’s Claude Opus 4.6 charges $5 and $25 per million tokens — making the Chinese alternatives 10–20 times cheaper for developers and startups.
Overall, OpenRouter’s total weekly token usage has climbed to 12.1 trillion, a 12.7-fold increase compared to a year ago. Founded by former OpenSea CTO Alex Atallah, OpenRouter now offers unified API access to more than 400 models from over 60 providers.
According to MIT Technology Review, this cost advantage is fueling widespread adoption of Chinese open-weight models even within U.S. startups. Venture capitalist Martin Casado of Andreessen Horowitz estimates that about 80% of young AI companies using open-source stacks are now running on Chinese-made models.
