Chinese-developed large language models have taken all five top spots in OpenRouter‘s global usage rankings for the first time, according to the platform’s July 2026 data.
Xiaomi’s MiMo-V2.5 ranked first by token volume, followed by models from DeepSeek, MiniMax, Alibaba’s Qwen family and Moonshot’s Kimi, according to OpenRouter.
OpenRouter processes more than 20 trillion tokens a week, and Chinese-origin models now account for more than 60% of all routed traffic, the platform’s data showed.
The shift has built through 2026. By April, Chinese models held six of the top nine spots, and by May they controlled four of the top five and about 61% of total token consumption, according to the source article.
CNBC reported in July that U.S. companies had shifted usage heavily toward Chinese models on OpenRouter because of cost advantages.
Jefferies global head of equity strategy Christopher Wood said on July 27 that cheaper Chinese open-source models could cause “massive capital destruction” in U.S. AI investments.
Wood has argued since early 2026 that China is better positioned to win the AI race because of advantages in energy supply and narrowing gaps in computing capability, according to the source article.
DeepSeek’s V4-Pro is priced at about one-twelfth the cost of OpenAI’s GPT-5.5 at comparable benchmark performance, while Xiaomi cut MiMo API prices by as much as 99% in May, the source article said.
Meta’s Llama, which helped define open-weight AI in 2023 and 2024, has fallen below 1% of routed volume, according to the source article.
Industry observers said routed inference volume does not capture the full market because Anthropic and OpenAI still command most industry revenue through premium enterprise pricing, while first-party consumer traffic through ChatGPT and Gemini remains far larger than router volumes.
OpenRouter’s data shows the open Chinese model tier has become the default choice for developers when cost is unconstrained by brand loyalty, according to the source article.
Featured image credit
