Alibaba’s Qwen family of open-weight AI models, rather than any American flagship, was the standout of the past year in artificial intelligence, according to a year-end column published by Wired. The piece, part of the magazine’s recurring “Expired, Tired, Wired” format, argues that the rise of Qwen and other Chinese open models has coincided with disappointing launches from OpenAI and Meta, and that the trend could reshape the industry’s balance of power in 2026.
The column’s central claim is that models should be judged less by benchmark scores than by how widely they are actually used. On that measure, Wired says, Qwen has been pulling ahead.
Adoption in real products
Wired points to Qwen’s presence in shipping consumer hardware, including smart glasses from a startup in Hangzhou, China, that use the model to translate spoken Mandarin into English text displayed on the lens in real time. The magazine presents this as the kind of practical deployment that American frontier models have been slower to reach.
The column also cites adoption by Western companies. Airbnb chief executive Brian Chesky said in October 2025 that his company relies heavily on Qwen because the models are fast and inexpensive, remarks that were widely reported at the time and later drew questions from members of the US Congress, according to Forbes. Wired adds that Nvidia, Perplexity and the Chinese automaker BYD have integrated Qwen models into products or systems, and that Meta has reportedly used Qwen in work on a new model of its own.
Download figures
According to Wired, data from the developer platform Hugging Face showed Chinese AI models overtaking American models in downloads for the first time in July 2025, with Qwen accounting for more than half of the Chinese share. The magazine puts cumulative Qwen downloads at more than 600 million. Forbes separately reported that the Qwen family had passed 700 million downloads on Hugging Face by January 2026, a figure consistent with the earlier count.
Wired attributes much of this traction to Qwen’s open-weight release model, which lets developers download, modify and fine-tune the models for their own needs without paying per-query fees to a provider. The column notes that the Qwen team also received a best paper award at the NeurIPS machine learning conference in December 2025.
American models under pressure
By contrast, Wired describes OpenAI’s GPT-5, released in August 2025, as having fallen short of the expectations built up before launch, with users and researchers criticizing its tone and pointing to basic errors. The magazine says Meta’s Llama 4 likewise failed to make a significant impact. Both assessments reflect Wired’s editorial judgment and the reactions it reports, rather than any measured comparison of model capability.
The column also credits DeepSeek, another Chinese developer, with adding to the pressure on US labs by showing strong performance at lower computing cost earlier in the year.
Wired’s conclusion is that if real-world usage is the yardstick, Qwen and other Chinese open models are gaining faster than their American counterparts, and that 2026 may show whether that momentum changes who sets the direction of the industry. A separate Wired report in the same period noted that Chinese labs are pitching their open models as stable, accessible alternatives as access to Anthropic’s and OpenAI’s frontier models becomes more restricted in some markets.
