On a drizzly On a windswept summer afternoon, I went to the Rokid headquarters, an innovative startup. smart glasses Hangzhou, China. While I was chatting with the engineers, they were quickly translated into English from Mandarin, then displayed on a tiny transparent screen, just above my left eye, using one of their new prototype devices.
Alibaba, the Chinese online giant that dominates Chinese commerce, developed Qwen – an open-weight model with large languages.
Qwen—full name 通义千问 or Tōngyì Qiānwèn in Chinese—is not the best AI model around. OpenAI’s GPT-5, Google’s Gemini 3Then, Anthropic’s Claude Often score higher than benchmarks that measure different aspects of machine intelligence. Qwen is not the first open-weight machine to be truly innovative. Meta’s LlamaThe social media giant released the 2023 version of its.
Yet Qwen, and other Chinese models—from DeepSeek, Moonshot AI, Z.ai, and MiniMax—are increasingly popular because they are both very good and very easy to tinker with. According to HuggingFaceIn July, the number of downloads on the platform of a Chinese AI model surpassed those of US models. DeepSeek made waves with its cutting-edge AI large language model, which required less computation than US counterparts. However, OpenRouter, an open platform that allows users to route queries through different AI models. says Qwen is the most popular model of the open type in the world.
Qwen is capable of doing most tasks you would expect from an AI. Rokid customers can identify items taken with a camera built into the device, get directions on a map or search the web. Rokid offers a customized version of Qwen, which can be easily downloaded and modified. You can also run a tiny version of Qwen in your smartphone or on other devices, just to be prepared for the event that internet connectivity fails.
Qwen is a great tool to have on your computer before you go. MacBook Air It was great for practicing some Mandarin basics. Open source, smaller models such as Qwen can be just as effective as those that reside in large data centres.
In the past 12 months, Qwen’s rise and that of other Chinese models with open weight have coincided in some cases with stumbles by famous American AI models. Meta released Llama 4 as a model in April of 2025. The performance, however, was disappointing, not reaching the levels set by popular benchmarks, such as LM Arena. This failure led many developers to look for alternative open-source models.
