The usage volume of China's large AI models has exceeded that of the United States for two consecutive weeks, and a mysterious model has entered the top ten.

Image source: Jiemian Image Library

On March 16th, the latest weekly data (from March 9th to March 15th) of OpenRouter, the world's largest AI model API aggregation platform, shows that the weekly call volume of Chinese large AI models has exceeded that of US models for two consecutive weeks. The former's weekly call volume rose to 4.69 trillion Tokens last week, while the latter's dropped to 3.294 trillion Tokens.

The top three models in terms of Token call volume last week were all Chinese large models, namely MiniMax M2.5, Step Star Step 3.5 Flash (free), and DeepSeek V3.2. Among them, MiniMax M2.5 had a weekly call volume of 1.75 trillion Tokens. Although it decreased by 6% compared with the previous week, it has ranked first for five consecutive weeks. Step Star Step 3.5 Flash (free) had a weekly call volume of 1.34 trillion Tokens, a significant increase of 79% compared with the previous week. DeepSeek V3.2 had a weekly call volume of 1.04 trillion Tokens, an increase of 25% compared with the previous week.

In addition, the weekly call volume of Kimi K2.5 of another domestic large model, Dark Side of the Moon, was 0.56 trillion Tokens, a decrease of 25% compared with the previous week, ranking ninth.

Among the top ten large models in terms of call volume last week, a mysterious model named "Hunter Alpha" newly entered the list, with a weekly call volume of 0.666 trillion Tokens, ranking seventh. The introduction shows that Hunter Alpha is a cutting - edge intelligent model with 1 trillion parameters and a 1 - million - Token context, specifically built for agent applications. It is good at long - term planning, complex reasoning, and continuous multi - step task execution, and has the reliability and instruction execution accuracy required by frameworks such as OpenClaw.

In the previous week (from March 2nd to March 8th), the weekly call volume of Chinese large models was 4.19 trillion Tokens, and that of US large models was 3.63 trillion Tokens. Among the top five large models of that week, MiniMax M2.5 had a weekly call volume of 1.87 trillion Tokens, ranking first; DeepSeek V3.2 had a weekly call volume of 0.83 trillion Tokens, ranking third; Step Star Step 3.5 Flash had a weekly call volume of 0.75 trillion Tokens, ranking fifth.

OpenRouter is the world's largest large - model API aggregation platform, which can provide developers with a unified API interface to access hundreds of large language models around the world. Its core functions include multi - model calls, intelligent routing optimization, and a transparent performance leaderboard, aiming to solve the problems of complex multi - model integration and vendor lock - in.

The platform's data shows that from February 9th to 15th, Chinese AI models' call volume reached 4.12 trillion Tokens, exceeding that of US models (2.94 trillion Tokens) for the first time. From February 16th to 22nd, the weekly call volume of Chinese models further climbed to 5.16 trillion Tokens, while the call volume of US models during the same period dropped to 2.7 trillion Tokens. Among the top five models in global call volume, MiniMax's M2.5, Dark Side of the Moon's Kimi K2.5, Zhipu's GLM - 5, and DeepSeek's V3.2 occupied four seats.

A recent research report by Dongguan Securities stated that as the programming and agent capabilities of domestic models improve, their call volume has increased significantly. Domestic large models in the programming and agent fields are comparable to global leading models, which is expected to further accelerate the implementation of applications and the growth of Token consumption.

This article is from Jiemian News, reporter: Chen Xiaotong. Republished by 36Kr with permission.