Two US-built artificial intelligence coding assistants, Cursor and Windsurf, recently announced the launch of their proprietary models, Composer and SWE-1.5, respectively. The rollout took an unexpected turn when users discovered that both tools were actually running on Chinese-made AI systems.
Developers in several countries began noticing Chinese-language text appearing in code snippets generated by Cursor. Around the same time, Windsurf confirmed that its core model was provided by Z.ai (formerly Zhipu AI) after the company’s official account on X reposted Windsurf’s launch announcement with a congratulatory note.
The revelation drew mixed reactions online, with some users joking that it might be time to start learning Mandarin. But beneath the humor lies a deeper story about how quickly China’s AI ecosystem is reshaping the global landscape.
In the US, relationships between AI model developers and application startups have long been fraught. When OpenAI was reportedly in talks to acquire Windsurf earlier this year, Anthropic immediately cut off Windsurf’s model access. Now, with open-source models climbing global leaderboards, developers are increasingly turning to non-US providers for flexibility and lower costs.
Windsurf isn’t alone. Vercel recently added Z.ai’s GLM-4.6 to its API offerings. Cerebras, a major US chipmaker, said it would begin promoting GLM-4.6 as a primary model starting November 5. Together AI has deployed Alibaba’s Qwen-3-Coder, while inference platform Featherless now supports Moonshot AI’s Kimi K2.
RELATED ARTICLE
Just a year ago, Chinese companies were integrating models from OpenAI and Anthropic into their systems. Now, the flow has reversed. The reason, analysts say, is straightforward: China’s open-source models have caught up in both performance and cost. The rise of DeepSeek, Alibaba’s Qwen, and the growing prominence of Z.ai have strengthened their credibility.
The fact that two of the most popular US coding assistants rely on Chinese models underscores the competitiveness of China’s AI developers, especially in coding, where benchmarks and performance are easy to quantify.
Affordability is another major factor. Social Capital founder Chamath Palihapitiya recently said Groq migrated workloads to Moonshot’s Kimi K2 model because it was “way more performant and frankly just a ton cheaper” than offerings from OpenAI and Anthropic. Cost efficiency, he added, has become the AI sector’s new frontier.
Coding tools are one of the few AI categories that have achieved clear product-market fit. Anthropic currently leads, with OpenAI’s GPT-5 and Codex expected to intensify competition. Meanwhile, Chinese firms are moving quickly to secure market share.
Data from OpenRouter, which aggregates over 100 AI models, shows that four of the five most used products as of September were coding-related tools. Within that group, Chinese models are rapidly gaining traction. According to PPIO, Z.ai’s GLM and Moonshot’s Kimi ranked among the most used globally in the third quarter, with GLM’s usage peaking above 10% and Kimi holding between 2–5% market share.
Chinese developers are also turning this momentum into direct revenue. In September, Z.ai introduced a subscription service for developers similar to Anthropic’s Claude Code, priced between RMB 20–200 (USD 2.8–28), roughly one-tenth the cost of its US counterpart. The plan supports more than ten major coding tools and is available worldwide. According to 36Kr, the service has already generated annual recurring revenue of over RMB 100 million (USD 14 million).
Following that launch, Moonshot rolled out its own developer package for Kimi on October 24, and MiniMax introduced a free limited version of its M2 model with plans for a paid tier.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Xiao Xi for 36Kr.