Very long, GPT-5. Hi Quinn

On rainy This summer afternoon, I visited the headquarters of Rokid, a development startup Smart glasses In Hangzhou, China. As I spoke with the engineers, their words were quickly translated from Mandarin to English, then transcribed onto a small, transparent screen above my right eye using one of the company’s new hardware prototypes.

Rokid’s high-tech glasses use Qwen, a large, open-weight language model developed by Chinese e-commerce giant Alibaba.

Qwen – full name 通义千问 or Tōngyì Qiānwèn in Chinese – is not the best AI model ever. OpenAI‘s GPT-5, Google‘s Gemini 3and Anthropic‘s Claude They often score higher on benchmarks designed to measure different dimensions of machine intelligence. Qwen is also not the first truly advanced openweight model dead‘s Llamawhich the social media giant launched in 2023.

However, Qwen and other Chinese models — from DeepSeek, Moonshot AI, Z.ai, and MiniMax — are increasingly popular because they are so good and easy to manipulate. According to HuggingFacea company that provides access to AI models and code, downloads of open Chinese models on its platform surpassed downloads of US models in July this year. DeepSeek has shaken the world by releasing a large, state-of-the-art language model with far less computation than competitors in the US, but OpenRouter, a platform that routes queries to various AI models, He says Qwen’s popularity rose rapidly over the year to become the second most popular open model in the world.

Qwen can do most of the things you want from an advanced AI model. For Rokid users, this might include identifying products captured by a built-in camera, getting directions from a map, drafting messages, searching the web, etc. Since Qwen can be easily downloaded and modified, Rokid hosts a copy of the model, fine-tuned to suit its purposes. It is also possible to run a small version of Qwen on smartphones or other devices just in case the Internet connection is interrupted.

Before going to China I installed a small version of Qwen on my device MacBook Air I used it to practice some basic Mandarin. For many purposes, modestly sized open source models like Qwen are only as good as the behemoths that live inside large data centers.

The rise of Qwen and other open-weight Chinese models has coincided with the stumbles of some popular US AI models in the past 12 months. When the Meta revealed Llama 4 in April 2025, the model’s performance was disappointing, failing to reach the top of popular benchmarks such as LM Arena. The voucher left many developers looking for other open models to play with.