Latest data shows that among the tokens generated by U.S. companies via model calls made through OpenRouter, about 58% flow to Chinese models—around three times that of mid-January this year.

This March, Chinese models’ share in U.S. corporate traffic first surpassed U.S. models for the first time; in the first week of July, that ratio briefly reached 63%, setting a new high.
Roll the timeline back to early 2025, when across the entire OpenRouter platform, Chinese models’ token share was still below 10%, while U.S. models accounted for about 80%. After one and a half years, Chinese models have already completed the overtaking.
As of the 13th of this month, DeepSeek has become the largest model developer by token usage across the entire OpenRouter platform. In the past six months, shares for Chinese vendors such as Xiaomi, MiniMax, and Tencent have also been rising.
Now look at pricing: at the cheapest service nodes at the time, DeepSeek V4 Flash costs only $0.09 and $0.18 per 1 million input and output tokens, respectively; GPT-5.5 is $5 and $30, respectively.
Model nationality is complicated—but for the boss, once they’ve seen the bills, it’s not.