‘Qwen has become part of the default workflow for developers’, Hugging Face said.

Alibaba’s open-weight models have garnered more than 3bn downloads globally over the past six months, surpassing offerings from US heavyweights Google, OpenAI, Meta and others despite the Chinese company’s considerably smaller investments.

Hugging Face, in a recent report analysing open models on its platform, found Qwen’s 2bn-plus downloads to be the most this year – by a large margin. Beijing Academy of Artificial Intelligence products ranked second at 519m, followed by models from Google, at 418m, and OpenAI, at 329m.

Alibaba revealed that its Qwen family of AI models have open-sourced more than 460 models that have led to more than 300,000 derivatives. More than half that figure sits on the Hugging Face hub, nearly five times Meta’s Llama footprint. Google follows with Alibaba with 82,506 derivatives.

“Qwen has become part of the default workflow for developers deciding what models to fine-tune and deploy,” Hugging Face said in its latest State of Open Models report, published on 14 August.

The open platform for AI datasets and tools explained that Alibaba’s consistent release schedule for its AI models, continuous updates, wide ranging sizes and use cases, and ease of model modification for commercial release have contributed to its chart-leading position.

According to an internal memo reviewed by news publications, the Chinese technology giant is now gearing up to sell its gaming arm Lingxi Games to Asian private equity firm Trustar Capital – reportedly for up to $1.5bn and to free up capital for further AI spending.

This is part of a broader corporate reorganisation at Alibaba, as it divests non-core assets to make AI and cloud computing its top priorities. The company intends to spend more than $55bn on AI over the next three years.

Chinese AI start-ups have launched several open models much larger than their US counterparts in 2026.

Last month, Moonshot AI’s Kimi K3 boasted 2.8trn parameters while promising capabilities close to OpenAI’s new GPT-5.6 Sol and Anthropic’s Fable 5, while Alibaba launched the 2.4trn-parameter Qwen3.8-Max, which ranks even higher than K3 on several benchmarks.

These new open models are also increasingly optimised for domestic chips in China, as US curbs on semiconductor exports to China over recent years have instead pushed the country to focus on its own chip capabilities.

US models stayed under 130bn parameters in five of seven months the Hugging Face report covered, with Nvidia’s Nemotron 3 Ultra and Thinking Machines Lab’s Inkling being the only exceptions.

“Building large stopped being a differentiator,” Hugging Face said in its report. Xiaomi and Meituan both cleared a trillion parameters this year, it explained, adding that the size profile is a “statement of intent rather than of capability”.

US open source, meanwhile, is growing in areas where smaller models from the likes of Google, Microsoft, IBM and OpenAI still get hundreds of millions of downloads annually.

The report noted that larger US models launched this year with more than 100bn parameters are built on top of Chinese models or leverage artifacts from Chinese labs, including Inkling and two Nemotron 3 models.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.