US firms shift routine AI work to cheaper Chinese open models

Chinese AI labs Moonshot, Z.ai and DeepSeek are winning business from US companies by undercutting American rivals on price, Fortune reported on 26 July 2026.

Fortune said the price gap is wide: Anthropic's Claude Fable 5 costs $50 per million output tokens, against $15 for Moonshot's Kimi K3, $4.40 for Z.ai's GLM-5.2 and about $0.87 for DeepSeek-V4-Pro.

Coinbase chief executive Brian Armstrong said in late June that the company nearly halved its internal AI bill by making Zhipu's GLM 5.2 and Moonshot's Kimi K2.7 Code the default models for engineers, routed through an internal gateway that assigns each task by type and cost. The change lifted cache hits from 5% to 60%, and 91% of engineers never reached their original usage limits, according to Armstrong. The savings came from shifting routine work such as code review and document summaries away from top Anthropic and OpenAI models, not from cutting quotas.

DoorDash co-founder Andy Fang said the company assigns "lower-level work" to Kimi for better quality at lower cost. The company's internal DashBench test found that pairing Kimi K2.6 as a scout with Claude Fable 5 as reviewer produced the best results, at $3.81 per pull request.

Bloomberg reported that Airbnb, DoorDash and Coinbase run the Chinese models on their own servers.

The cost pressure comes as Moonshot expands its model line-up. The company released Kimi K3 on 16 July, which it describes as the largest open model ever built, with open weights due to follow on 27 July.

The shift has drawn scrutiny in Washington. The House Homeland Security Committee and the House Select Committee on China wrote to DoorDash asking for a list of the Chinese models it uses and details of its security testing, with a response due by 14 August.

US companies face no ban on using Chinese models, though some government agencies have prohibited the use of DeepSeek. DoorDash's response to the congressional request, due next month, will show how much detail the company is willing to disclose about its use of the models.