What happened
Reuters reported, citing two people familiar with Alibaba's plans, that the company intends to ask large commercial users of its next Qwen model, Qwen3.8-Max, for a share of the revenue they make from it. Qwen3.8-Max is an open-weight model, meaning the underlying trained settings are free to download and run on a company's own infrastructure. Until now, Alibaba has charged developers only for using Qwen through its own cloud hosting service, while generally letting businesses run the open-weight version in their own data centers at no cost. The reported change follows a path Moonshot, a rival Chinese AI lab, already set with its Kimi K3 model. Kimi K3's published license requires any company that resells the model as a service and earns more than $20 million a year doing so to negotiate a separate commercial agreement, which sources told Reuters can include a revenue share of up to 30 percent. A Chinese IT services provider, Chinasoft International, already disclosed a revenue-sharing agreement with Moonshot in a regulatory filing, without stating the percentage. Alibaba's own revenue-share rate has not been made public and, per Reuters, is still being worked out. The terms are expected to firm up around Qwen3.8-Max's release, reported to be as soon as next week. Reuters also quotes on-record sources describing this as a deliberate, common strategy: offer the model free to build adoption, then charge once a business is generating real revenue from it. The CEO of DigitalOcean, which already has a commercial agreement with Moonshot, called it a "tried and tested open-source freemium model."
Why it matters for business owners
Open source and open weight get marketed as the answer to vendor lock-in: no subscription, no per-token bill, run it yourself. That framing was true on day one of these Chinese open-weight releases. It is proving less true a few months later, once the labs that built the models need a return on the cost of training them. This matters most if your business, or a software vendor you depend on, has built a customer-facing product on top of an open-weight model like Qwen or Kimi K3 and now generates meaningful revenue from it. That business may be about to receive a phone call about commercial terms it did not budget for. It matters somewhat less, but is still worth noting, if you are running an open-weight model internally for your own workflows without reselling access to it, since the reported threshold targets large-scale commercial resale, not internal use. Either way, the underlying lesson is the same: a model's license terms today are not a permanent contract. They can change once the vendor decides the free tier has done its job.
What owners should not misunderstand
Do not read this as Alibaba, or open-weight AI generally, becoming closed source or unusable for free. Qwen3.8-Max will still be downloadable and runnable without payment for most uses. The reported change targets a specific category: large companies reselling the model as a hosted service and making significant revenue doing so. If that is not your business, the direct impact is limited for now. Do not assume this means Alibaba's terms will match Moonshot's. Reuters is explicit that Alibaba's percentage is undisclosed and still under discussion. The 30 percent figure and $20 million threshold belong to Moonshot's Kimi K3 license, cited here as a confirmed, named comparison, not as a preview of what Alibaba will actually charge. Do not treat this as unique to Chinese AI labs, either. It is the same freemium pattern used across software for decades: give away the product to build adoption, then monetize once usage is real. The news here is that it has now clearly arrived in open-weight AI licensing, and it arrived faster than many businesses assumed it would.
The operational lesson
A model being free to download is not the same as a model being free to build a business on. Those are two different commitments, and open-weight AI licenses are increasingly drawing a line between them: personal use and small-scale deployment stay free, while resale at scale becomes a negotiated commercial relationship. That line can move. Vendors set the free tier's boundaries, and they can redraw them as their own business model matures, exactly as Moonshot did with Kimi K3 and Alibaba is reportedly about to do with Qwen. A business that adopted an open-weight model assuming the license terms were fixed at the moment of adoption is treating a moving target as a stationary one. The practical fix is not to avoid open-weight models. It is to read the actual license, not the marketing, before betting a product on one, and to revisit that license periodically rather than assuming it still says what it said when you signed on.
What a serious business should do next
If your business, or a vendor you rely on, resells access to or builds a commercial product around an open-weight AI model, read the current license text directly rather than relying on how the model was described at launch. Note any revenue threshold, resale clause, or requirement to negotiate a separate commercial agreement once usage crosses a certain size. If you are evaluating Qwen3.8-Max, Kimi K3, or any other open-weight model for a product you plan to monetize, ask the vendor directly what usage level triggers a commercial negotiation and what that negotiation has historically cost other customers, rather than assuming the free tier scales indefinitely. Build a light habit of checking license terms on any open-weight model your business depends on, on a recurring basis, the same way you would review a cloud vendor's pricing page. A model's status as "open source" describes how it is distributed today. It does not guarantee what a vendor can ask for tomorrow. Keep your architecture flexible enough that a licensing change on one model does not force an emergency rebuild. That usually means knowing, in advance, which parts of your stack would need to change if a given model's terms shifted, and roughly what switching to an alternative would cost.
The Atlacis view
"Open source" and "free" get used interchangeably in AI marketing, and this story is a clean example of why that substitution is not safe to make long term. Alibaba has not done anything unusual here. It has done what most software vendors eventually do once a free tier proves its value: start charging the users making real money from it. Atlacis helps owners look past a model's license badge to the actual terms underneath it, map what a licensing change would cost their specific stack, and keep enough architectural flexibility that a vendor's pricing decision does not become an emergency. That kind of review belongs in the plan before a product is built on a model, not after a bill arrives.
The short version
- On August 7, 2026, Reuters reported that Alibaba plans to require large commercial users of its next Qwen open-weight model, Qwen3.8-Max, to share a portion of the revenue they generate from it, with terms expected to take shape around the model's release as soon as next week.
- This mirrors a provision already confirmed in rival Moonshot's Kimi K3 open-weight license, which requires companies reselling the model at scale and earning more than $20 million a year from it to negotiate a commercial agreement, reported at up to a 30 percent revenue share. Alibaba's own rate is undisclosed and still under discussion.
- The change targets large-scale commercial resale, not personal use, research, or most internal business deployment. The model stays free to download for most users.
- The underlying lesson is broader than either company: open-weight AI license terms are not fixed at the moment a business adopts a model. Vendors can, and increasingly do, redraw the line between free and paid once usage is real.
- Any business building a commercial product on an open-weight model should read the current license directly, track revenue or usage thresholds that trigger a commercial negotiation, and keep enough architectural flexibility to absorb a licensing change without an emergency rebuild.