A new artificial intelligence model named Qwen is rapidly increasing in popularity among developers and companies worldwide. Developed by Chinese e-commerce giant Alibaba, Qwen offers a flexible open-weight model that allows for easy modification and integration. This rise in adoption highlights a shift in the global AI landscape.
Despite not always topping benchmark scores, Qwen's accessibility and adaptability are proving to be key factors in its growing success. It represents a different approach to AI development compared to some of its larger, more closed counterparts.
Key Takeaways
- Qwen, an open-weight AI model from Alibaba, is seeing rapid global adoption.
- Its ease of modification and integration appeals to developers and companies.
- Chinese open-weight models have surpassed US models in downloads on platforms like HuggingFace.
- Major companies, including BYD, Airbnb, and Nvidia, are integrating Qwen into their products.
- The model's openness contrasts with the more closed approach of some US tech giants.
The Rise of Open-Weight AI Models
The AI community is seeing a growing interest in open-weight models. Unlike proprietary systems, open-weight models allow developers to download, inspect, and modify their code. This transparency fosters innovation and allows for greater customization.
Qwen, also known as 通义千问 (Tōngyì Qiānwèn), is a significant player in this open-weight movement. While models like OpenAI’s GPT-5 or Google’s Gemini 3 may achieve higher scores on certain benchmarks, Qwen’s practical utility and ease of use are driving its widespread adoption.
Did You Know?
Downloads of open Chinese AI models on HuggingFace surpassed those of US models in July of this year, indicating a significant shift in developer preference.
Practical Applications of Qwen
Companies are already integrating Qwen into diverse products. For instance, Rokid, a startup based in Hangzhou, China, uses Qwen in its smart glasses. These glasses provide real-time language translation and transcription directly onto a translucent screen above the user's eye.
Rokid's engineers have fine-tuned a version of Qwen to suit their specific needs. This customization allows the smart glasses to perform tasks such as identifying products, providing directions, drafting messages, and searching the web. The ability to run a smaller version of Qwen on devices like smartphones also ensures functionality even without an internet connection.
"A lot of scientists are using Qwen because it's the best open-weight model," says Andy Konwinski, cofounder of the Laude Institute, an organization advocating for open US models.
Chinese Innovation vs. US Strategies
The success of Qwen and other Chinese models, including those from DeepSeek, Moonshot AI, Z.ai, and MiniMax, highlights a difference in development philosophies. Chinese AI companies often prioritize publishing detailed papers on their engineering and training methods. This openness allows the broader scientific community to learn from and build upon their work.
This approach stands in contrast to the increasingly closed nature of some major US AI companies. These firms often guard their intellectual property closely, making it harder for external developers to understand and adapt their models. This difference in strategy may be contributing to the growing popularity of open Chinese models.
About Open-Weight Models
Open-weight models, sometimes referred to as open-source in the AI community, make their underlying code and trained parameters publicly available. This allows developers to download, modify, and redistribute the models, fostering collaboration and rapid iteration.
Impact on the Global Market
The rising prominence of Qwen suggests that a model's true value extends beyond raw benchmark scores. Its utility in enabling other innovations, its ease of integration, and its transparent development process are proving to be powerful advantages.
Companies like BYD, China’s leading EV maker, have integrated Qwen into new dashboard assistants. Beyond China, US firms are also adopting Qwen. Airbnb, Perplexity, and Nvidia are all reportedly using Qwen. Even Meta, a pioneer of open models with its Llama series, is said to be utilizing Qwen in the development of new models.
Academic Recognition
Hundreds of academic papers presented at NeurIPS, a leading AI conference, have utilized Qwen. A paper from the Qwen team detailing a method to enhance model intelligence during training was recognized as one of the best papers at the conference this year.
Challenges for Established AI Models
The past year has seen some notable AI models face challenges. Meta's Llama 4, unveiled in April 2025, reportedly disappointed many developers with its performance, failing to meet high benchmarks. Similarly, OpenAI's GPT-5, released in August, received mixed feedback, with some users noting simple errors and an unusually cold demeanor.
While OpenAI did release a less powerful open model called gpt-oss, Qwen and other Chinese models continue to gain traction. This is partly due to the continuous effort put into building and updating these models, alongside the widespread publication of their engineering details.
The focus on narrow benchmarks, measuring specific skills like math or coding, might be leading some US AI companies astray. Andy Konwinski suggests that these benchmarks may not always reflect real-world usage or problem-solving needs. This creates a disconnect between development goals and practical impact.
Ultimately, the measure of an AI model's success may increasingly hinge on how widely it is used to build other applications and drive real-world innovation. By this metric, Qwen and its open Chinese counterparts are showing strong momentum.