Alibaba Releases Qwen3.8-Max with Revenue Share Terms
Alibaba launches Qwen3.8-Max on Hugging Face, imposing revenue-sharing terms for large enterprises mirroring Moonshot AI.
Stock photo for illustration only, not from the actual event
- Alibaba uploaded Qwen3.8-Max model weights to Hugging Face on August 12, 2026
- Businesses generating over $50 million must secure paid commercial licenses
- Beijing-based Moonshot AI previously introduced a similar 30% revenue share for Kimi K3
- The new policy introduces strategic risks and shifts in the open-weight AI landscape
Alibaba has released the model weights for Qwen3.8-Max, its most advanced AI model, on Hugging Face on August 12, 2026, allowing entities to deploy it within their own data centers. However, this release introduces unprecedented commercial conditions compared to previous Qwen models. Businesses that utilize the model to provide services and hit specific revenue thresholds are now required to enter into commercial agreements and share a portion of their revenue back with Alibaba.
Sources familiar with the company's plans indicate that Alibaba aims to extract revenue shares from major users of Qwen3.8-Max. Although exact rates remain under negotiation with individual partners, the strategic direction is clear: freely downloadable models are beginning to carry financial strings attached.
The published Hugging Face artifact is designated as Qwen3.8-2.4T-A95B, featuring a total of 2.4 trillion parameters while activating only 95 billion parameters per inference pass via a Mixture-of-Experts (MoE) architecture. Stripped of vision capabilities to focus purely on text, the model requires a data center setup with BF16 weights totaling roughly 4.89 terabytes and deployment on NVIDIA GB300 NVL72 hardware containing 72 GPUs per rack, targeting cloud providers and large enterprises directly.
Stock photo for illustration only, not from the actual event
Alibaba's approach is not entirely unprecedented in China, as Beijing startup Moonshot AI embedded similar terms into the license for Kimi K3 upon its release in mid-July 2026. Under Moonshot's terms, entities offering Model as a Service (MaaS) with over $20 million in annual revenue must execute separate commercial contracts, while products exceeding 100 million monthly active users (MAU) or $20 million in monthly revenue must visibly display the Kimi K3 brand.
Sources report that Moonshot's revenue cut reaches up to 30%, with concrete deals already materializing. Chinese IT services firm Chinasoft International announced to the Hong Kong Stock Exchange in July that it formed a token-based revenue-sharing agreement with Moonshot to build AI agent solutions for energy, power, and financial sectors, sending Chinasoft shares up over 30% in a single day.
Alibaba's Qwen3.8-Max license follows this exact framework while raising the bar. Businesses offering model services or AI assistants with trailing 12-month revenues exceeding $50 million must purchase a paid commercial license. Internal enterprise usage without exposing capabilities to third parties remains free, and branding requirements are triggered at 100 million monthly users or $20 million in monthly revenue.
"We pay to work with the labs making these open-weight models, to make sure the deployment is optimized, and to get early access to the next generation of models."
Paddy Srinivasan, Chief Executive Officer of DigitalOcean Holdings
The transition from pure open-source accessibility to conditional revenue-sharing models highlights how top AI labs are striving for sustainable monetization after heavy infrastructure investments. This freemium-style approach allows companies to rapidly capture user adoption before extracting value from large commercial clients, though it introduces new competitive dynamics against alternative models that remain entirely free.
Source: Techsauce
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment