Global AI large-model usage hits 44.6 trillion tokens, and Chinese models have topped the charts for seven consecutive weeks

June 8 to 14, global AI large model token call volume surged to 44.6 trillion, with Chinese models accounting for 18.42 trillion for seven consecutive weeks topping the charts, with the top four spots all held by Chinese models such as DeepSeek, MiniMax, Tencent, and others.
(Previous summary: OpenRouter analyzes 100 trillion tokens in research reports: What are humans really using AI for, the rise of Chinese models, and user retention secrets)
(Additional background: 2026 AI model ultimate rankings: Who is the strongest among Claude, GPT-5.5, Grok, Gemini?)

Table of Contents

Toggle

• All top four spots are held by Chinese models
• The rapid rise of MiniMax M3 is remarkable
• Chinese AI model market share continues to expand

The weekend just passed, from June 8 to 14, the global AI large models on the OpenRouter platform reached 44.6 trillion token calls, a 23.5% increase from the previous week, marking the eighth consecutive week of growth. Chinese AI models captured 18.42 trillion tokens during the same period, a 29.8% increase month-over-month, accounting for nearly 40% of the global total; American models reached 5.72 trillion, with a 78.7% increase but a lower base. Chinese large models have called more than American models for seven consecutive weeks, maintaining the global lead.

All top four spots are held by Chinese models

Among the top five in OpenRouter call volume last week, the top four were all Chinese AI large models, with specific rankings as follows:

• DeepSeek-V4-Flash: with 4.41 trillion tokens, ranked first for four consecutive weeks, a 20% increase MoM
• MiniMax M3: 4.32 trillion tokens, rose to second place, a 73% surge MoM
• Tencent Hy3 Preview: 4.14 trillion tokens, remained third for consecutive weeks, a 41% increase MoM
• The fourth spot was occupied by a Chinese model

Notably, DeepSeek V3.2, which ranked ninth last week, dropped out of the top five, marking its first absence since December of last year.

The rapid rise of MiniMax M3 is remarkable

MiniMax M3 has skyrocketed from outside the rankings to second place within just a few weeks, with a 73% MoM growth rate, second only to the US model group. This coincides with the time when MiniMax officially open-sourced the 428B native multimodal MoE model at the beginning of June—after open-sourcing, developers can access it at zero cost, greatly boosting call volume.

Chinese AI model market share continues to expand

Based on OpenRouter data, the token call volume share of Chinese models has increased from about 30% in early March to nearly 40% now. Including long-tail Chinese models not listed, the actual share could be even higher.

The driving force behind this trend is the price competitiveness of Chinese AI models. Models like DeepSeek series, MiniMax, Tencent Hy3, and others generally have API prices 50% to 80% lower than comparable US models, attracting large-scale migration by developers. Although US models grew by 78.75% this week, mainly driven by explosive growth in high-priced models like GPT-5.5, Claude 4 Opus, and others, the base effect has not yet reversed the overall landscape.