Alibaba Plans to Charge Revenue-Sharing Fees for Heavy Users of Qwen Open-Source Models
Alina Collins
Alibaba will embed revenue-sharing terms in its upcoming open-source model Qwen3.8-Max, charging larger commercial users a cut of their earnings. China's AI giants are collectively pivoting from free open-source to a freemium model — free to use, pay when you profit.
How exactly will Alibaba charge?
Qwen3.8-Max remains open-source and open-weight — developers can download the parameters and deploy for free.
But Alibaba will add revenue-sharing clauses to the license, targeting commercial users above a certain annual revenue threshold.
This means → small teams and solo developers stay free; only companies that make serious money with the model trigger the fee.
In plain terms = the model itself costs nothing, but once your business built on it reaches a certain scale, Alibaba takes a cut.
Didn't Moonshot's Kimi K3 already do this?
Kimi K3, released last month by Chinese startup Moonshot, already includes similar terms: any entity earning over $20 million annually from selling services based on the model must sign a commercial agreement.
Sources say Moonshot's revenue-share rate runs as high as 30%.
Chinasoft International (HKEx: 0354) disclosed a revenue-sharing deal with Moonshot in regulatory filings but did not reveal the exact percentage.
This reflects a broader pattern — Alibaba is not experimenting alone; China's AI industry is collectively testing this commercial path.
Do Chinese models still have a price edge?
Kimi K3's per-token pricing is roughly one-third that of Anthropic's Fable model.
This means → Chinese models remain far cheaper than U.S. peers; the revenue share is an additional monetization layer on top of low pricing.
In plain terms = hook users with low prices first, then charge the heavy commercial users — the classic "cheap acquisition, deep monetization" playbook.
What does this mean for developers and enterprise users?
Previously, Alibaba's open-source models could be deployed free in customers' own data centers. The new terms narrow that window for large-scale commercial users.
Dan Fu, VP of Kernels at Together AI, noted that AI service providers profit by optimizing token efficiency — "At the application layer, how you use the model and make tokens truly valuable — that's where the business opportunity lives."
Lin Qiao, CEO of Silicon Valley-based Fireworks AI, said he sees no fundamental barrier to powerful open-source U.S. models: "We are really waiting for that day to come."
This signals that Silicon Valley is watching China's monetization moves closely. If the model proves out, U.S. open-source models will very likely adopt similar terms.
Content is for reference only, not financial advice.