Similar to last month’s major release by the Chinese AI startup Moonshot, Alibaba’s Qwen3.8-Max model is an open-source, open-weight system. This means that developers can download the foundational settings needed to implement or modify the model. In contrast, U.S. developers OpenAI, Anthropic, and Google’s Alphabet use closed-source models.
Open source is often perceived as free; however, last month’s Kimi K3 release by Moonshot included a clause requiring anyone selling the model as a service with annual sales exceeding $20 million to negotiate a commercial agreement with Moonshot.
Alibaba intends to introduce a similar provision for its open-source model next week, according to the two sources, who requested anonymity due to the confidential nature of the plans.
Historically, Alibaba has charged developers to utilize its models on its own cloud computing platform while allowing most of its open-source products to be employed in clients’ data centers at no cost.
This strategy indicates that Chinese AI companies, which have surprised markets by unveiling open-source models nearly as advanced as those from OpenAI and Anthropic, are moving toward a unified business model to capture market share from their U.S. competitors.
REVENUE-SHARING DEALS
In emerging agreements, U.S. companies are expected to share revenue earned from Chinese models with the Chinese AI labs responsible for their creation, even as the U.S. government has accused Moonshot of technology theft from Anthropic, a claim that Chinese officials have dismissed.
According to the two sources, Moonshot’s Kimi K3 license mandates its partners to share revenue, with Moonshot seeking up to a 30% revenue share. Alibaba also plans to request a revenue share, but the specific rate remains uncertain as negotiations are ongoing.
Moonshot did not respond to requests for comment.
Last month, Chinasoft International, a Chinese IT services firm, revealed a revenue-sharing agreement with Moonshot in a regulatory document, though the percentage was not specified.
The Chinese AI firms are adopting a familiar strategy from Silicon Valley: initially offering software at little to no cost, followed by charges for extensive commercial usage and additional services.
“You invest in collaboration with these open-weight model labs to ensure you are optimizing your deployment. You pay for early access to the next version of the model,” stated Paddy Srinivasan, CEO of DigitalOcean Holdings, a U.S. company offering Kimi K3 and other Chinese models.
He confirmed that his company has a commercial agreement with Moonshot, but he refrained from discussing specific details.
“This follows a proven open-source ‘freemium’ model,” Srinivasan noted.
Kimi K3 is priced about one-third lower than Anthropic’s Fable model, based on the listed costs of input and output tokens for both models.
Dan Fu, vice president of kernels at Together AI, explained that companies in the AI software space profit by optimizing services, like improving token efficiency, which are the essential components of AI queries.
“At the application layer, there’s significant value in how you utilize it, including how you obtain the models and tokens for practical applications,” Fu elaborated.
Meanwhile, open-source AI continues to grow, with U.S. companies entering the market.
Thinking Machines Lab, a San Francisco-based AI startup founded last year by former OpenAI Chief Technology Officer Mira Murati, launched its first open-source model last month and is expected to release more advanced models soon.
“I don’t see any fundamental obstacles” to the development of powerful open-source U.S. models, remarked Lin Qiao, CEO and co-founder of Silicon Valley-based Fireworks AI, who chose not to disclose his company’s business dealings with Moonshot. “We are just waiting for that to unfold.”
