Chinese models account for the most usage on OpenRouter when measured in tokens

If you measure AI usage in tokens, Chinese models take up the most space on the OpenRouter platform. That is shown by the platform's own leaderboard, with data through November 7, 2026.

A token is a small chunk of text that an AI model reads or writes. The more tokens, the more work the model has done. OpenRouter is a service that lets developers use many different AI models through a single access point.

DeepSeek leads the pack

On the leaderboard, a DeepSeek model, V4.1 Flash 1017, is the most used, with about 56 trillion tokens in a week, an increase of 17 percent. In second place is DeepSeek V4.1 Flash with about 17.7 trillion tokens, an increase of 21 percent.

Together that comes to about 73.7 trillion tokens. Based on the figures for the ten models shown, that equals roughly 56 percent of the ten most-used models' combined total of about 132 trillion tokens for the week. The share is calculated from the leaderboard's figures.

The other models in the top 10 are GLM 5.3 Flash from Z.ai (13.5 trillion), Tencent's new Hy4 (10.1), OpenAI's GPT-6 Luna (7.26, down 29 percent), Step 5 Preview from StepFun (6.99), Anthropic's Claude Opus 5.5 (6.74), MiniMax M3.1 Flash (5.37), Tencent's Hy4 Preview (4.2, down 55 percent) and Kimi K3.1 Pro from Moonshot AI (4.12, up 109 percent). Only OpenAI and Anthropic are US providers; the rest are Chinese.

Agents drive the growth

In an analysis, OpenRouter writes that DeepSeek had slightly under 10 percent of weekly token flow in January 2026, and that its share fell to about 5 percent in February and March, when the use of AI agents rose sharply. An AI agent is a program that solves tasks on its own in multiple steps, for example by using tools.

According to OpenRouter, DeepSeek reached nearly 20 percent in June and became the most-used model in mid-May. Its share has roughly doubled in six months. At the end of May, V4 Flash made up 70 percent of DeepSeek's agent-based token flow, one month after its release.

OpenRouter estimates that agent-based work uses about 15 times more tokens per request than human use. That can produce large volume without necessarily meaning more users.

Caveats

Token share is not the same as revenue or number of users. The leaderboard only measures traffic through OpenRouter and is not a picture of the entire AI market. Figures for other periods are based on different weeks and definitions and cannot be compared directly.