US firms quietly switch to cheaper Chinese AI models

What happened: US companies are shifting routine AI workloads onto Chinese open-weight models, Fortune reported on 26 July. Coinbase, DoorDash and Airbnb are among the adopters.

The numbers: Per million output tokens, Anthropic's Claude Fable 5 costs $50, Kimi K3 costs $15, Z.ai's GLM-5.2 costs $4.40, and DeepSeek-V4-Pro costs about $0.87.

Who's affected: Coinbase made Zhipu's GLM 5.2 and Moonshot's Kimi K2.7 Code the defaults for engineers, routed through an internal gateway by task and price. Caching hit rates rose from 5% to 60%, and 91% of engineers never hit their original usage limits. Savings came from moving code review and document summaries off top Anthropic and OpenAI models, not from cutting quotas.

In their words: DoorDash co-founder Andy Fang said the company hands "lower-level work" to Kimi for better quality at lower cost. Its DashBench test found Kimi K2.6 as a scout paired with Claude Fable 5 as reviewer produced the best results, at $3.81 per pull request.

The context: The Decoder reported that Airbnb, DoorDash and Coinbase all run Chinese models on their own servers. Moonshot released Kimi K3 on 16 July, calling it the largest open model ever, with open weights due 27 July.

The catch: US firms aren't banned from using Chinese models, though some government agencies have banned DeepSeek. The House Homeland Security and China committees have written to DoorDash asking for a list of Chinese models it uses and details of its security testing, due 14 August.

Why it matters: Cheap, open-weight Chinese models are undercutting Western labs on price for everyday tasks, forcing companies to reconsider which AI model handles what.

What's next: DoorDash's response to Congress, due 14 August, will show how companies balance cost savings against political pressure over Chinese AI adoption.