[American Companies Allocate 60% of Tokens to Domestic Models, Price Advantage Translates into Usage Advantage] According to a report by Jinse Finance, as of July 21, OpenRouter data shows that the proportion of Tokens used by American companies for Chinese models has risen from less than 10% a year ago to nearly 60% by early July, peaking at 63%. DeepSeek has been a major driver of this trend. This indicates that domestic models are no longer just being tested by developers but are now integrated into the actual workflows of American companies, leveraging their low cost, open weights, and sufficiently strong programming and agent capabilities. However, this does not mean that domestic models have captured 60% of the U.S. corporate AI market. The statistics in the chart reflect the number of Tokens on the OpenRouter platform and do not include traffic directly routed through providers like OpenAI or Anthropic. Additionally, lower-cost models are more likely to handle large-scale, long-chain tasks, which naturally amplifies their Token share. This also suggests that companies are reserving high-end models for complex tasks while delegating a large volume of standardized work to domestic models that are "good enough in capability and lower in cost." AI competition is shifting from single-model rankings to multi-model orchestration and per-task cost efficiency. (Jin10)