Chinese AI Models Consume 18.8 Trillion Tokens in Week Ending June 22, Surpassing U.S. Models

According to Jefferies' AI report published on June 22, Chinese AI models consumed 18.8 trillion tokens in the week ending June 22, surpassing U.S. models at 5.8 trillion. DeepSeek V4 Flash ranked first with 4.94 trillion tokens, followed by Xiaomi's MiMo-V2.5, MiniMax M3, and Tencent's Qwen. OpenRouter data showed platform-wide token consumption grew 4.7% week-over-week to 46.7 trillion.

The shift reflects Chinese models' competitive balance between performance and cost. Jefferies noted Chinese models now narrowed the intelligence gap with U.S. counterparts while offering API costs at a fraction of American alternatives, attributed to MoE architecture and optimized attention mechanisms. Enterprise spending remained subdued, with Jefferies' LLM Token Expenditure Index at 1.64–1.68 on June 14–19, down from 2.04 on May 31, indicating developers shifted toward cheaper, more efficient models.