Claim: no Qwen commercial-serving licenses granted since Qwen 3.8

The open-weights corner of X is debating whether the commercial-serving door is closing. A single post claims Alibaba has granted no commercial-serving licenses since Qwen 3.8 shipped roughly two months ago; the checkable fact underneath is that the Qwen 3.8 license text sits on Hugging Face where anyone can read it.

Key facts

• @0xkydo claims no commercial-serving licenses have been granted since the Qwen 3.8 release ~2 months ago — unverified.
• The same post claims similar restriction patterns for MiniMax M3.1 Flash — unverified.
• The Qwen license files on Hugging Face are primary sources, readable directly.

What's driving the conversation

The @0xkydo post argues that since the Qwen 3.8 release about two months ago, no commercial-serving licenses appear to have been granted to inference providers, and that MiniMax's M3.1 Flash shows a similar pattern of revenue and attribution limits. Medium velocity, but the claim lands at the exact intersection of this week's two license-relevant stories: MiniMax M3.1 Flash's community-license weights and the unanswered license question on Mistral Large 4's pending release.

Replies so far are operators comparing notes on which models their providers still list, plus the usual skepticism that a single account can know the state of Alibaba's enterprise licensing pipeline. Nobody has produced a license grant or a rejection letter.

The substance

The verifiable layer: Qwen's model cards on Hugging Face carry their license files in the open, and provider model listings show who is serving what right now. A systematic check — current license texts, provider listings, any Alibaba licensing announcements — would confirm or falsify the no-grants claim without needing Alibaba's cooperation. That check has not been published. What stays claim-only: the grant-cessation assertion itself, which is one account's inference from silence, not a documented policy change.

Why builders are watching

If confirmed, this is the most consequential quiet shift in open-weights serving: labs keeping "open weights" branding while withholding commercial-serving grants would mean hosted providers and decentralized networks can no longer legally serve the most popular models, pushing builders toward Apache-2.0 alternatives like Nanbeige's 4.3-33B. It extends the tightening pattern DeAI documented in Qwen's image-model license change and gives concrete meaning to the open-weights versus open-source distinction. Watch Mistral's Large 4 license terms this week — they will show whether the pattern crosses from Chinese labs to European ones.