Over the past two months, China has seen a flurry of new frontier AI models released, including Moonshot AI’s Kimi K3, Z.AI’s GLM-5.3-Flash, Alibaba’s Qwen 3.8-Max, and DeepSeek’s DeepSeek-V4-Pro.
However, when examining the latest technical benchmark rankings, a few recent business developments reveal even more about the evolving landscape of Chinese AI technology.
A US-based legal AI firm supported by OpenAI, Sequoia Capital, and Andreessen Horowitz introduced its inaugural in-house model, which was trained further on Kimi K3. The following day, Thomson Reuters announced its own proprietary model, developed at an estimated cost of $40 million utilizing Qwen. Just last week, Alibaba launched an international version of its AI assistant platform, QwenWork, opening it up to users outside of China.
These aren’t groundbreaking breakthroughs; many US companies are building on Chinese foundations—Cursor’s coding model is based on Kimi K2.5, Airbnb relies on Qwen, and Perplexity is built on DeepSeek. In June, Lindy transitioned from Anthropic to DeepSeek. Several advanced tech companies in the U.S. are effectively constructing their products on Chinese-developed models.
This trend persists because AI in China is characterized by open weights, affordability, and application-driven design. Companies generate revenue through related services, but generally, models can be downloaded and operated on local servers. Due to efficiency improvements—partly driven by US export restrictions—Chinese models can deliver high performance at lower computational costs. Meanwhile, many top US AI models require metered access, resulting in much higher expenses for token usage.
In essence, Chinese firms are providing high-quality AI solutions at a fraction of the cost, with open weights allowing for more customization and flexibility.
A key factor behind this divergence is that US research labs have publicly stated they aim to develop artificial general intelligence. In contrast, Chinese companies tend to focus on creating models designed to be integrated into various economic applications. For example, Alibaba’s Chairman emphasized the importance of open source because it enables a broad spectrum of companies and entrepreneurs to innovate in AI, rather than concentrating ownership among a few giants.
My company, WPIC Marketing + Technologies, exemplifies this approach. We assist global consumer brands in selling across China, Japan, South Korea, and Southeast Asia by managing their online stores, logistics, warehouses, and marketing efforts. We employ several hundred staff, primarily based in Nanjing, Beijing, and Hangzhou.
When we first incorporated AI into our operations, using top US frontier models was not cost-effective, as much of the efficiency gained would have been captured upstream. The open-weight models allowed us to download Qwen and Kimi, deploy them on our own infrastructure, and customize them extensively using 15 years of China e-commerce data. This enabled us to develop a comprehensive AI suite, called Webber, which includes analytics, creative production, logistics management, and more—all hosted on Chinese open models running on our servers. The efficiency improvements have been substantial.
China’s approach to AI dissemination offers additional advantages. Both government and large corporations have taken proactive steps to address potential social harms associated with AI. Courts in China have ruled that companies cannot dismiss employees solely to replace them with AI. JD.com has committed to retraining warehouse workers as automation advances. New regulations restrict AI companion services for minors. At the recent World AI Conference in Shanghai, President Xi Jinping emphasized that AI should serve humanity.
Aligned with this ethos, I personally assured WPIC staff earlier this year that no one would be laid off due to AI adoption.
However, this doesn’t mean disruption will be prevented entirely. Some companies might promise not to lay off workers but then do so quietly. Automation will phase out certain job categories, as in other parts of the world, and retraining programs are often easier to announce than effectively implement.
Notably, Stanford’s 2026 AI Index reports that 83% of people in China view AI as more beneficial than harmful, compared to only 39% of Americans. The Chinese population appears more receptive to AI technologies, and global companies are increasingly relying on China’s leading models to support core functions and innovations.
China’s open, low-cost, application-focused AI ecosystem—bolstered by safeguards against negative externalities—demonstrates significant strength in this rapidly evolving field.