TL;DR

Alibaba intends to require a share of revenue from major commercial users of its forthcoming Qwen3.8-Max open-weight model, on the account of two sources with knowledge of the strategy, and the measure is due next week. It follows Moonshot, whose Kimi K3 licence obliges anyone reselling the model as a service above $20m in annual sales to negotiate commercial terms — reportedly seeking as much as 30%. Neither rate is a download fee; both are levies on the businesses built on top.

Free to run, chargeable to resell

Until now Alibaba has charged for model use on its own cloud and let most open-weight releases run in customers’ data centres without payment. The change does not close that door so much as put a meter on the large commercial end of it. Chinasoft International used a regulatory filing last month to disclose that it shares revenue with Moonshot, without naming the percentage.

The pattern is a familiar Silicon Valley one — give the software away, charge for heavy commercial use and the services around it. DigitalOcean chief executive Paddy Srinivasan, whose firm offers Kimi K3, described what the payment actually buys: collaboration on deployment optimisation and early access to the next revision. He confirmed a commercial agreement with Moonshot exists, called the structure a tried and tested freemium model, and declined to give terms.

What this changes for UK buyers

Any UK business whose AI cost model assumes free inference on open weights should read the licence rather than the announcement. The distinction that matters is between running a model internally and building a product on it that others pay for — the second is what these terms target, and £20m-scale UK software firms selling AI-backed services are inside the range being described.

The price gap driving adoption remains substantial: Kimi K3 runs at roughly a third of Anthropic’s Fable on listed input and output token prices. That gap is what made open Chinese weights the default cost-reduction move for firms squeezed by frontier pricing, and a revenue share narrows it without closing it.

Looking forward

Open-weight releases are no longer only a Chinese story. Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, shipped its first open model last month. Fireworks AI chief executive Lin Qiao sees no fundamental barrier to capable US open models arriving. Whether those follow the same licensing route will determine if revenue-sharing becomes the standard commercial shape of open weights, or stays a Chinese-lab convention.