The Rise of Qwen: How Chinese Open-Weight AI Models Are Gaining Global Prominence

This summer, at Rokid’s headquarters in Hangzhou, China, the company demonstrated its smart glasses technology powered by Qwen, an open-weight large language model developed by Alibaba. The glasses translate Mandarin to English in real-time and display the results on a tiny screen above the eye.

Why Qwen and Chinese Models Matter

While Qwen may not outperform top models like GPT-5 or Claude on technical benchmarks, it has become increasingly popular for its combination of capability and accessibility. According to HuggingFace, downloads of open Chinese models surpassed US ones in July of this year, with Qwen rapidly becoming the second-most-popular open model globally.

The appeal lies in Qwen’s versatility and adaptability. It can identify products through cameras, provide directions, draft messages, search the web, and more. Since it can be easily downloaded and modified, companies like Rokid can fine-tune it for specific applications. Even a small version can run locally on devices like smartphones when internet connectivity is unavailable.

The Openness Advantage

Chinese AI models are gaining traction partly due to stumbles from American counterparts. Meta’s Llama 4 disappointed in April 2025, failing to reach expected benchmarks. Similarly, OpenAI’s GPT-5 release in August underwhelmed users who reported coldness in responses and surprising errors.

The contrast between Chinese and American approaches to AI development is striking. While US companies increasingly protect their intellectual property, Chinese AI companies routinely publish papers detailing engineering and training techniques. A paper from the Qwen team was even recognized as one of the best at the premier AI conference NeurIPS this year.

Growing Adoption

Both Chinese and American companies are integrating Qwen into their products. BYD, China’s leading EV manufacturer, has incorporated the model into a new dashboard assistant. US companies like Airbnb, Perplexity, and Nvidia are also using Qwen. Even Meta, once the pioneer of open models, is reportedly using Qwen to help build a new model.

Hundreds of academic papers presented at NeurIPS used Qwen, with Andy Konwinski of the Laude Institute noting that “a lot of scientists are using Qwen because it’s the best open-weight model.”

A New Benchmark for Success

The rise of Qwen suggests that a key measure for AI models should extend beyond technical benchmarks to include how widely they’re used to build other products and services. Konwinski argues that US AI companies have become too focused on gaining marginal edges on narrow benchmarks at the expense of real-world impact.

By the metric of practical application and adoption, Qwen and other open Chinese models are clearly ascending in the global AI landscape.