/ The open-source AI power balance shifted.
Liam Ortiz is a tech journalist who covers AI related big tech and breaking news at TheTweaks. Before joining TheTweaks he worked for almost four years in corporate and national tech news in different companies. Few are quick but Liam is quicker, he breaks news before anyone else and that makes her special. His passion is somewhere connected with profession as his hobby is watching documentary movies.
There is a number circulating this week that most readers will breeze past 3 billion downloads. But pause a moment to reconsider, because this single number completely recontextualizes the entire open source AI race.
Alibaba’s Qwen AI models have now surpassed 3 billion total downloads in only 6 months. By way of comparison, in 2026 Google’s open models had 418 million downloads, while Meta’s Llama family of open source AI models got 227 million downloads, according to the recently published “State of Open Models” report by Hugging Face. Sum up the numbers of Google and Meta and Qwen outperforms them by about 4.6 times. That is not an advantage. That is a different league. (Bloomberg)
Most stories that cover this achievement focus on the number of downloads. This is an obvious take and an easy story. A much more interesting question is: Why are developers, not just in China but worldwide, opting for Qwen AI models more often as the choice of model that needs to be fine tuned, deployed and built on?
According to the “State of Open Models” report of Hugging Face: Qwen has essentially become a part of the default workflow of developers choosing what to fine-tune and deploy. This difference is significant compared to the total volume of downloads that could have been boosted by curious looks or one-time experiments. Being somebody’s default means that Qwen is winning the boring and tedious battle of fine tunes, enterprise deployments and derivative models that keep getting built on top of it.
Alibaba has made more than 460 Qwen models open source and has reported that developers use Qwen models as the base to build 300,000 derivative models. That is an ecosystem rather than a single event of a spike in the volume of downloads.
And there is a key aspect that most of the stories miss: the Qwen downloads are happening not in a void. Alibaba Cloud has been successfully distributing Qwen AI models among enterprise customers in Southeast Asia and Africa, the regions where Meta and Google do not have strong cloud infrastructure and enterprise sales departments.
This is a huge distribution advantage that most of the “open source AI models” comparisons do not consider. Being on the top of the downloads ranking in Hugging Face is one thing. Having an active cloud sales department promoting your models for enterprise contracts is quite another. Alibaba Cloud is doing both at once.
The thing that is worth noting here, since credibility is more important than virality: the 3-billion figure is provided by Alibaba Company itself, in their press release. By contrast, the figures of Google and Meta downloads are cited directly from the independent “State of Open Models” report by Hugging Face. This is not a reason to discount Alibaba’s claims. The general trend of line is clearly visible by now: derivative models’ counts, cloud distribution strategy, developers’ sentiment – all those trends point in one direction. However, it is important to know that you are comparing a self reported number against two independent sources’ numbers. Readers deserve to know this context before accepting “3 billion” as a hard fact.
Download and derivative models’ counts have become the only tangible metric that allows measuring the winner of the broader US-China AI competition, especially in the open weight models sphere, when the access itself is the product. Every derivative model that gets built on Qwen is a small vote of confidence, some developer decided that Qwen’s architecture, license terms and performance was better than Llama or Gemini open models’.
Western AI companies have noticed. Just recently Meta and Nvidia have released new open source models, a response which appears to be less of coincidence and more of a struggle to retain their position before Qwen’s dominance becomes permanent rather than temporary.
Alibaba is not planning to stop there. Their next big release, Qwen3.8-Max is a 2.4 trillion parameters model designed specifically for coding and agentic tasks, the areas of current concentration of enterprise AI investments. According to the rumors, Alibaba Cloud has been exploring a revenue sharing arrangement for their biggest commercial customers using Qwen models through Alibaba Cloud. It appears that the era of “open source as goodwill” for Qwen is about to end and “open source as a monetized funnel” is coming.
If the trend will develop, the download numbers will stop being vanity metrics and become a business model – a business model that Meta and Google, still treating open models as purely ecosystem plays, not as a monetized funnel, have not yet matched.
The 3 billion number is real, impressive and self reported all three statements are compatible with each other and none of them invalidates the other. But the key thing to understand here is not the round number itself, but the underlying trend: Qwen is not just winning downloads, it is winning the repeat business of the open source AI developer community, thanks to the cloud distribution strategy that Meta and Google have not yet matched in other regions beyond their home turf.
It is the combination of volume, stickiness and distribution that makes this milestone interesting to pay attention to. Our read: this is not just a one time victory of Alibaba and a hint of future development in 2026.
For more AI and tech news, visit TheTweaks AI News or check out our latest tech launches and big tech coverage.
Tim Cook’s 15 years as Apple’s chief executive officially ended on September 1 and now the torch has been passed on to John Ternus, who is the company’s long-time hardware. It’s Apple’s first leadership change…
Be respectful and constructive. Have a question or feedback? We’d love to hear from you. Contact us at contact@thetweaks.com