As the prices of leading U.S. large model services continue to rise, more and more American tech companies are quietly integrating Chinese open-source AI models into their production-level infrastructures. Cryptocurrency exchange Coinbase recently took the lead in making adjustments, successfully reducing AI costs by switching models without restricting employee usage.

Cost Reduction Becomes the Core Driving Force

Coinbase CEO Brian Armstrong revealed that the company has set Zhipu GLM 5.2 and Moonshot Kimi K2.7 as the default large models for all engineers. This move is mainly used for routine tasks such as code reviews and document summaries. Under the condition of an exponential increase in token usage, it has helped the company cut AI expenses by nearly half.

Since most engineers had never reached the previous usage limits, this optimization not only did not reduce employees' token quotas but also proved the universality of this approach. In addition to Coinbase, well-known companies such as Airbnb and AI company Lindy have also recently switched to different Chinese open-source models due to cost pressures.

Chinese Open-Source Models Gain Popularity Overseas

Industry estimates show that several Chinese open-source large models can achieve performance comparable to leading overseas models at significantly lower prices. On the well-known AI model calling platform OpenRouter, Chinese large models have long been among the top-tier text models in terms of calls.

Currently, multiple Chinese models, including DeepSeek, Xiaomi MiMo, MiniMax, Tencent Huan Yuan, and Zhipu GLM, are winning favor in the overseas market with their high cost-effectiveness. This dual advantage of technology and cost is quietly reshaping the AI infrastructure layout of global technology companies.