Lindy, a U.S. startup utilized by over 400,000 professionals, has replaced its primary artificial intelligence model from Anthropic’s Claude Sonnet to DeepSeek’s v4 Flash.
The Cost and Performance Shift in Global AI Models
Affordable AI models developed outside the United States are increasingly challenging Western counterparts on both price and capability. According to benchmark evaluations published by Artificial Analysis, Anthropic’s Fable 5.1 achieved an intelligence score of 57, but Chinese competitors Moonshot AI’s Kimi K3 and Zhipu AI’s GLM 5.3 closely followed with scores of 50 and 49, respectively. These Chinese models outscored Google’s Gemini 3.8 Flash (rated at 47) while operating at a fraction of the cost.
Task-specific pricing underscores the widening gap. Artificial Analysis data shows that running a standard benchmark task on Anthropic’s Fable 5 model costs approximately $5.62, compared to $1.58 for Moonshot AI’s Kimi K3 and $1.26 for Zhipu AI’s GLM 5.3. This cost efficiency has driven rapid adoption across international developer platforms. On OpenRouter, a major AI model aggregation platform, the token usage market share for Chinese models—including DeepSeek, Zhipu AI, Alibaba, and Tencent—reached 43.9% by the end of last month, surpassing U.S. models at 42.8%. One year prior, Chinese model usage stood at 14.6%.
Enterprise Adoption Beyond Startups
Major global enterprises are integrating Chinese AI architectures into their core customer service and industrial operations. Airbnb utilizes Alibaba’s Qwen models for customer support applications, while Siemens employs DeepSeek for administrative and industrial tasks. Airbnb Chief Executive Officer Brian Chesky noted that the Qwen architecture offers high performance, speed, and cost-effectiveness.
Capital markets are reflecting this commercial momentum. Zhipu AI cleared its initial public offering (IPO) hearing on the Hong Kong Stock Exchange on the 19th of last month, marking the first time a Chinese AI firm cleared the exchange’s listing review. Competitor MiniMax followed with its own IPO approval days later on the 21st, while industry peers Evoken and Moonshot AI initiated similar listing preparations on the Hong Kong bourse.
Open-Weight Strategies and Global Expansion
Unlike many proprietary U.S. frontier systems, several prominent Chinese AI developers utilize an “open-weight” distribution model. By releasing core model weights, these companies allow international enterprises to download, modify, and run the software locally on their own servers rather than relying on external cloud APIs.
This deployment method addresses corporate data sovereignty concerns. When U.S. lawmakers raised questions regarding enterprise use of overseas architectures, Airbnb clarified that it is not an Alibaba customer and does not provide data to Chinese companies. Similarly, telecommunications provider Indosat in Indonesia selected DeepSeek models for national agricultural and educational infrastructure projects, pointing to the flexibility of localized deployment.
According to a report titled “Token Politics” published by the Center for Strategic and International Studies (CSIS) last month, open-weight models serve as a significant software diplomacy tool for extending technical standards and developer community influence across emerging markets.
Keep reading