Alibaba's Open-Source AI Isn't Really Free Anymore
Qwen built its reputation on open weights — now Alibaba wants the companies profiting most from it to pay up.
Written by OutOfToken AI
August 10, 2026 · 4 min read · Synthesized from reporting by AI News · How this works
Alibaba is preparing to reshape the economics of its Qwen open-weight AI models, according to Reuters, which cited two people familiar with the plans. The next generation of Qwen will reportedly carry revenue-sharing terms for certain commercial users, marking a departure from the fully permissive open-source approach that helped Qwen become one of the most widely adopted model families outside the US.
Who Actually Pays
The arrangement, as described to Reuters, wouldn't target hobbyists or small developers building on Qwen. Instead, it's aimed at larger companies that turn Qwen into a paid service — think cloud providers, AI platforms, or enterprises reselling access to the model's capabilities. Those companies would need to strike a commercial agreement with Alibaba rather than simply download the weights and monetize freely.
The Missing Number
Alibaba hasn't finalized or disclosed the exact revenue-sharing percentage, and it remains unclear how the company would define thresholds for which businesses qualify as large enough commercial users to trigger the requirement. That ambiguity is likely to worry developers who've built products on the assumption that Qwen's licensing terms would stay static.
"Qwen has functioned as one of open-source AI's biggest success stories — until now, without a toll booth."
Why Now
Training frontier-scale models is enormously expensive, and Alibaba has continued shipping increasingly capable Qwen releases, including the recently unveiled Qwen3.5, which arrives in both open-weight and hosted API form with expanded agentic capabilities. Open-sourcing flagship models has given Alibaba enormous reach and developer mindshare, particularly in markets wary of US-based AI providers, but that reach hasn't translated directly into revenue the way a hosted API business would. A revenue-sharing layer lets Alibaba keep the openness that fueled Qwen's adoption while still capturing value from the companies making the most money off it.
A Broader Pattern
Alibaba has already been restructuring how it organizes AI internally, folding Qwen and related efforts into a new business unit, Alibaba Token Hub, alongside its Tongyi Large Model group and other AI initiatives under CEO Eddie Wu. That consolidation suggests Alibaba is treating its AI portfolio less as a research showcase and more as a coordinated commercial operation — one where licensing terms, monetization, and infrastructure are managed centrally rather than left to goodwill and community adoption.
If Alibaba follows through, it would test whether 'open-weight' and 'commercially free' can keep meaning the same thing at frontier scale. Other model makers weighing similar moves will be watching closely to see whether developers and enterprise users accept the tradeoff — or route around it.
Editorial Note
The research corroborates the core claims about Alibaba's revenue-sharing model for commercial Qwen users, the lack of finalized rates, and details about Qwen3.5. However, the sources are primarily social media posts and corporate pages that appear to cite the same Reuters report rather than independent verification. The research does not address some contextual claims about Qwen's broader success or market adoption outside the US.
Claim Tracker
AI-assessed
Source 1 (AI News) and Source 2 (Instagram) both confirm this reporting from Reuters citing two people familiar with the plans.
Source 1 explicitly states this; Source 3 further specifies 'major commercial users of Qwen3.8-Max' would need to share revenue.
Source 1 states 'Alibaba hasn't finalized or disclosed the exact revenue-sharing percentage'; Source 3 confirms the rate details remain undisclosed.
Source 6 (CNBC) confirms Qwen3.5 comes in open-weight and hosted API versions and supports new agentic capabilities.
The provided research confirms the policy targets 'larger companies' and 'major commercial users' but does not explicitly detail exemptions for hobbyists or small developers.
Ask AI about this story
// discussion
sign in to join the discussion