Key facts
- Alibaba plans to implement a revenue-sharing model for its upcoming Qwen3.8-Max open-source AI model.
- This model follows a similar strategy to Moonshot's Kimi K3, which requires commercial agreements for high-revenue users.
- Alibaba's move aims to capture market share from U.S. AI rivals by monetizing heavy commercial use of its open-source offerings.
- Previously, Alibaba charged for models on its cloud but allowed free use in customer data centers.
- Chinese AI firms are adopting a 'freemium' model, offering initial access at low or no cost and charging for extensive commercial use and advanced services.
Chinese technology giant Alibaba is preparing to charge major users of the next version of its Qwen open-source AI model a share of the revenue they generate from it, according to sources familiar with the company's plans. This strategy mirrors that of Chinese AI startup Moonshot, which has a similar provision in its Kimi K3 model's licensing terms.
Alibaba's Qwen3.8-Max model, like Moonshot's Kimi K3, is an open-source, open-weight model, meaning its underlying settings are available for download. This contrasts with closed-source models from U.S. developers like OpenAI, Anthropic, and Google.
Moonshot's license requires companies generating over $20 million in annual sales to negotiate a commercial agreement, with one source indicating a potential revenue share of up to 30%. Alibaba plans a similar revenue-sharing approach, though the specific rate is still under discussion. Chinasoft International has already disclosed a revenue-sharing deal with Moonshot.
This move by Chinese AI firms signals a convergence toward a business model that offers initial access at low or no cost, with charges for heavy commercial use and additional services, a strategy often seen in Silicon Valley. Companies like DigitalOcean Holdings are already offering Chinese models under commercial agreements.
The open-source AI landscape is also seeing increased participation from U.S. firms, with Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, having released its first open-source model.