Qwen, the artificial intelligence model developed by Alibaba, is believed to have captured more than half of global downloads of open-source AI models. Researchers say the model is approaching nearly 1 billion downloads as of March, outpacing rivals including Meta’s Llama and models from DeepSeek.
According to US-based AI analytics firm Interconnect AI, cumulative downloads for Qwen reached 942.1 million as of March, equivalent to over 50% of all open-source AI model downloads worldwide.
In February 2026 alone, Qwen recorded 153.6 million downloads—more than double the combined downloads of the eight largest rival models, including Meta, DeepSeek and OpenAI. In the same month, total downloads of US-origin models were around 56 million.
Open-source AI models are typically distributed with open structure, source code and weights, allowing users to download, modify and reuse them. Alibaba has also released the source code for the latest Qwen 3.5 version, and the company has said the model’s capabilities could rival leading US systems such as OpenAI or Anthropic.
More than 100,000 derivative models built on Qwen are available on the open-source Hugging Face platform, reflecting rapid expansion by the global developer community through additional applications and new model development.
A report from the US-China Economic and Security Review Commission says Chinese firms and research institutions can share, modify and fine-tune platform models with limited upfront training costs. The report links this approach to a faster innovation cycle and to efforts to address computational resource constraints amid US tightening export controls on advanced semiconductors.
The development of AI agents this year is also highlighting the advantages of open-source models. Agent systems often require deep access to email, calendars and personal data, which can demand high customization and offline operation.
While closed-source models such as those from OpenAI or Anthropic do not allow core-structure interventions, open-source models permit customization and offline operation, making them more suitable for AI-agent development.
The rapid growth of Qwen is prompting adjustments from US technology companies. Google has released the open-source Gemma 4 to maintain its developer ecosystem, separate from its commercially focused Gemini line. NVIDIA has also introduced the open Nemotron model along with the NemoClaw agent platform, aiming to accelerate participation in the open-source AI ecosystem.
