American programming products output "Chinese"

Recently, Cursor and Windsurf, the programming tools in the US that were the first to achieve Product-Market Fit (PMF), have respectively launched their own models, Composer and SWE-1.5. Subsequently, it was confirmed that both models are based on Chinese large models.

Users from various parts of the world found that Chinese characters frequently appeared in the code output by Cursor. When directly asked, Windsurf clearly stated that its model is provided by Chinese "Zhipu AI".

Zhipu's official Twitter account also retweeted the news to offer congratulations.

This phenomenon quickly sparked heated discussions among overseas netizens. Some even joked, "Should we start learning Chinese?"

This situation is not unexpected. In the US, there has always been a delicate relationship between model companies and application enterprises.

This year, when OpenAI planned to acquire Windsurf, Anthropic immediately cut off its services to Windsurf. Now, as more and more open-source models are catching up with closed-source models in the rankings, application companies are of course the first to welcome them.

This is not just an isolated case.

Vercel, a US platform valued at $9.3 billion, announced that it would integrate Zhipu's GLM-4.6 into its official API services. Cerebras, the world's largest inference chip company, recently sent an official email stating that it would promote GLM-4.6 as its main model starting from November 5th. The cloud service platform Together AI announced the deployment of Alibaba's Qwen-3-Coder, and the AI inference platform Featherless also announced its support for the new model of Kimi K2 upon its release.

Why are overseas products increasingly "speaking Chinese"?

01

Role Reversal: US Tech Companies Start Using Chinese Large Models

A year ago, Chinese tech companies were still striving to access top US large models such as GPT and Claude.

Now, the trend has reversed: US AI companies are starting to adopt large models developed in China.

Actually, the reason is quite simple. Chinese open-source models are excellent, fast, and inexpensive.

Since July this year, domestic large model companies have been dominating various model rankings. This can be seen from the enthusiastic praise of GLM-4.6 by the boss of Vercel, as well as the frequent appearances and popularity overseas of models like DeepSeek, Qwen, and Kimi-K2 since the beginning of this year.

The choice of domestic large models as the core engines by top overseas programming tools like Cursor and Windsurf once again proves that domestic large models have global competitiveness, especially in the Coding field, where they have been recognized by core players.

The second factor is cost-effectiveness.

In an interview, Chamath Palihapitiya, the founder of Social Capital, said bluntly, "We have started using Kimi-K2 on Groq. Although the models of OpenAI and Anthropic perform well, they are simply too expensive."

Overall, cost-effectiveness has become the key reason for US AI companies to choose to deploy Chinese large models. Chinese AI manufacturers are quite familiar with price competition.

02

Globalization: AI Coding is an Unavoidable Battlefield

In fact, it's not just the global market that needs Chinese large models. Globalization is also a battle that Chinese large model enterprises must win.

AI Coding is the first track in the AI field to achieve PMF in the past two years and is also the most fiercely competitive track in globalization. In this field, Anthropic from overseas is leading the way, and OpenAI's release of GPT-5 and Codex products is considered a response to Anthropic's technical approach.

Domestic large model manufacturers still in the game have also "coincidentally" set their sights on the Coding battlefield, which is a certain area. OpenRouter, a global large model supermarket that connects hundreds of large models, reflects the true preferences of global developers through the model call volume on its platform and is currently the "filter" closest to the truth. In September 2025, among the top 10 products in terms of call volume on OpenRouter, 4 out of the top 5 were Coding tools.

These Coding tools are also embracing Chinese large models, which has led to a change where Chinese large models, which used to have little presence on OpenRouter, quickly dominated the rankings after their release. According to PPIO data, the biggest change on OpenRouter in the third quarter was the greater variety of domestic model options. Among them, Zhipu's GLM and Kimi were the most popular among users. Kimi accounted for about 2% - 5% of the market share, and the usage of GLM once exceeded 10%.

In addition to making inroads in overseas providers and on OpenRouter, domestic large models are also actively exploring new commercial opportunities in Coding. In September this year, Zhipu was the first to launch a monthly subscription package supporting Claude Code. The package is priced between 20 and 200 yuan, which is about 1/10 of the price of Claude. It supports access to more than 10 mainstream programming tools and is being promoted globally. According to exclusive information from "Intelligent Emergence", Zhipu's Coding package has achieved good revenue two months after its release, with an estimated annual recurring revenue (ARR) of over 100 million yuan.

Subsequently, Kimi also launched a similar discounted monthly subscription package for programmers on October 24th, which is Programmers' Day. MiniMax M2 was released with a limited-time free offer and announced that it would also launch a Coding package soon.

In today's highly competitive large model market, the key to breaking the deadlock lies in choosing a widely verified scenario and making a bet. This bet is not only a test of perseverance but also a test of intelligence and judgment.