This follows a broader pivot Alibaba already began in 2026: transitioning earlier flagship Qwen models (Qwen3.6-Max, Qwen3.7-Plus) from open-source to API-only, closed models with usage fees, as commercialization pressure mounted . Qwen3.8-Max marks a return to open-weight release, but now with strings attached.
The Qwen3.8-Max model itself is a technical powerhouse: 2.4 trillion total parameters, a sparse mixture-of-experts architecture that activates only about 95 billion parameters per operation, a 1-million-token context window, and benchmark claims that put it alongside Anthropic's Fable and OpenAI's ChatGPT . Alibaba also matched US closed-model API pricing ($2 per million input tokens, $6 per million output tokens) .
But the strategic story is the licensing. According to Reuters, citing two people familiar with the company's plans, Alibaba will require major users of Qwen3.8-Max to share a portion of their revenue . The exact revenue-sharing rate and revenue threshold have not yet been publicly defined .
The move mirrors a licensing approach already adopted by Chinese AI startup Moonshot AI for its Kimi K3 model . Moonshot published its weights under a custom "kimi-k3" license on July 27, 2026, which is nearly MIT for basic use but contains two commercial gates .
| Feature | Alibaba Qwen3.8-Max (reported) | Moonshot Kimi K3 (published license) |
|---|---|---|
| License type | Custom, not yet disclosed | Custom "kimi-k3" license |
| Threshold for large users | Not yet specified (reportedly "major" commercial users) | $20M+ aggregate revenue over any consecutive 12 months from MaaS business |
| Revenue share | Reported revenue-sharing deal required for large users; rate not finalized | Up to 30% revenue share, according to at least one source |
| Attribution mandate | Not yet known | Products with >100M MAU or >$20M monthly revenue must prominently display "Powered by Kimi K3" |
| MaaS permission | Not yet known | Large MaaS providers need separate permission from Moonshot |
| Small users | Free to use | Free to use, no reporting required |
The key difference: Moonshot's license is already published and explicit (you can read the exact thresholds on GitHub), whereas Alibaba's license terms have not yet been disclosed as of August 7, 2026—the weights drop is expected within days . This uncertainty has made developers wary .
Alibaba (Qwen3.8-Max) — based on Reuters reporting :
Moonshot (Kimi K3) — based on the published license text :
Several forces are driving this monetization push:
Commercialization pressure: Alibaba (and other Chinese AI labs) have invested billions into training frontier models with no clear return yet. The shift to API-only on earlier Qwen models and now revenue-sharing on Qwen3.8-Max is a direct response to the need to monetize .
Race with US rivals: Qwen3.8-Max is positioned directly against Anthropic's Fable and OpenAI's ChatGPT, with Chinese companies claiming to close the capability gap rapidly . This is part of a broader "open model blitz" by Chinese AI firms that has alarmed US model makers .
Emulation of Moonshot's precedent: Moonshot's Kimi K3 was the first major Chinese open-weight model to introduce a revenue-triggered commercial tier. Alibaba is now following the same template, and the industry is watching whether this becomes the new norm for Chinese open-weight AI .
US-China tech rivalry: US export controls on advanced semiconductors have not stopped Chinese labs from producing frontier-class models. Chinese firms are leveraging open-weight distribution as a strategic counterweight to US closed-model dominance, gaining global developer adoption while seeking new monetization paths .
Uncertainty in the market: Developers are cautious. After Kimi K3's surprise revenue-sharing clause and MiniMax H3's geo-restrictions, the community is watching the Qwen3.8-Max license announcement—expected within days—with caution .