Alibaba plans to require large commercial users of the next version of its Qwen open-source AI model to share a portion of revenue generated from services built on the model, according to people briefed on the company’s plans. The move marks a push to monetize open-weight models while expanding Alibaba’s global AI strategy.
The model, reported to be Qwen3.8-Max, will be released with weights available for download, meaning developers can run or adapt the system on their own infrastructure. That open-weight approach contrasts with the closed-source models from U.S. companies such as OpenAI, Anthropic and Google. Sources said Alibaba intends to roll out revenue-sharing terms as early as next week, though the exact percentage has not been finalized.
The strategy mirrors licensing language introduced by Chinese startup Moonshot for its Kimi K3 model, which requires commercial partners who generate more than $20 million in annual sales from the model to negotiate commercial terms. Reports have said Moonshot’s agreements can include revenue shares of up to about 30%.
Until now, Alibaba charged customers for access to its models when they are hosted on Alibaba Cloud, but it has generally allowed users to deploy many of its open-source releases inside their own data centers without payment. The planned change signals a broader shift among Chinese AI labs toward a “freemium” playbook: make powerful models widely available to spur adoption, then monetize heavy commercial usage or value-added services such as optimization, hosting and early access to updates.
Industry participants say that firms offering AI infrastructure and services typically earn revenue by improving how models are deployed — reducing token costs, optimizing throughput and integrating models into applications. DigitalOcean’s CEO, for example, said marketplaces or service providers often strike commercial deals with model creators to secure collaboration and technical advantages.
The emergence of revenue-sharing agreements is already visible in regulatory filings: Chinasoft International disclosed a commercial arrangement with Moonshot last month, though it did not reveal the revenue-share percentage. The broader pattern has regulators and industry watchers tracking how open-source releases will be commercialized and how licensing terms will shape market access.
U.S. policymakers recently accused Moonshot of misappropriating technology from Anthropic, an allegation Chinese officials have rejected. At the same time, U.S. startups are also entering the open-source arena. Thinking Machines Lab — founded by a former OpenAI chief technology officer — released an open-source model this year and is expected to announce more powerful models in time.
Observers note price differences across models: listed token prices suggest Kimi K3 is substantially cheaper than some alternatives, such as Anthropic’s Fable, which has helped drive demand for Chinese open-source models among cloud providers and application developers.
Alibaba’s pending revenue-share plan underscores a growing convergence in commercial strategy among AI labs: open-sourcing advanced models to accelerate adoption, while preserving paths to monetize substantial commercial deployments. The exact terms and implementation approach for Alibaba’s Qwen3.8-Max remain subject to internal discussion and have not been publicly disclosed.
