Systems Research系统研究Oct 17, 20262026年10月17日Note讯息

Open weights you can't ship are a trial, not an exit.不能上线的开放权重，是试用，不是出口。

An empty road at blue hour leads through a wide open entrance into a walled courtyard; across it, the only exit booth glows amber with its barrier arm lowered, and a small figure stands looking at it.蓝调时刻，一条空路穿过敞开的大门进入围墙内的广场；广场另一端唯一的出口岗亭亮着琥珀色灯光，栏杆放下，一个小小的身影望着它。AI-generated image · concept, not evidenceAI 生成图像 · 概念示意，不作为事实证据

The default read of a free download is "no lock-in." Qwen's newest image weights move the lock-in from the API bill to the license page.

On Oct 9, Qwen published Qwen-Image-2.1-Turbo on Hugging Face: a 7B checkpoint that runs in 8 denoising steps. It ships under the Qwen Research License, dated Sept 20, 2026. Commercial use needs a separate license. The agreement is governed by Chinese law, with Hangzhou courts, and a breach lets Qwen terminate and require you to delete the weights.

That is a change of direction. The original Qwen-Image, in August 2025, shipped under Apache-2.0. Qwen's text models still do: Qwen3.8-27B, from August 2026, is Apache-2.0. Meanwhile Alibaba Cloud's Model Studio lists a hosted qwen-image-3.0-pro.

The mechanism: the open checkpoint becomes the place to try, and the hosted model becomes the place to ship. Exposure widens. The exit cost moves into the license.

Second-order effect: forks blur the terms. On Hugging Face's trending text-to-image list, read Oct 17, two community quantized copies of the Turbo model showed about 295,000 and 328,000 downloads, against 17,421 for the official repo. One of them is tagged Apache-2.0.

A fork's license tag is metadata. The base model's license is the contract, and the team that ships carries the risk.

The strongest counter: switching is cheap. Apache-2.0 image models exist, from FLUX.1-schnell to several released this month. For drafts, mockups and anything where good enough is the bar, the research license costs you nothing.

It holds less when one model's quality is the product. The new permissive entrants on the same list show tens to about a thousand downloads, a weak sign of ecosystem support, since downloads are not quality.

If you use open image models in paid work: before wiring one in, open the base model's LICENSE file, not the fork's tag. Then name the permissive model you would switch to, and run your real prompts on it once. This is not legal advice.

This breaks if Qwen relicenses the 2.1 family as Apache-2.0, or publishes a self-serve commercial license, by Jan 31, 2027.

看到免费下载，默认的读法是“没有锁定”。Qwen 最新的图像权重，却把锁定从 API 账单挪到了许可协议上。

10 月 9 日，Qwen 在 Hugging Face 发布了 Qwen-Image-2.1-Turbo：一个 70 亿参数、8 步去噪即可出图的模型。它采用 2026 年 9 月 20 日发布的 Qwen Research License。商业使用需要另签商业许可；协议适用中国法律、由杭州的法院管辖；一旦违约，Qwen 可以终止授权并要求你删除权重。

这是方向上的变化。2025 年 8 月的初代 Qwen-Image 用的是 Apache-2.0。Qwen 的文本模型至今仍是：2026 年 8 月发布的 Qwen3.8-27B 采用 Apache-2.0。与此同时，阿里云 Model Studio 的模型列表里有一个托管的 qwen-image-3.0-pro。

机制是这样的：开放的权重变成了“试”的地方，托管模型变成了“上线”的地方。你能接触的范围变宽了，退出成本却转进了许可协议。对个人开发者来说，这意味着你可以免费评估它的效果；可一旦要拿它赚钱，就得回到厂商那里谈条件，或者改用它的托管接口按量付费。

二阶效应：分支版本会把条款搞模糊。在 Hugging Face 文生图热门榜上（10 月 17 日读取），Turbo 模型的两个社区量化版本分别约有 29.5 万和 32.8 万次下载，官方仓库是 17,421 次。其中一个被标成了 Apache-2.0。

分支上的许可标签只是元数据，基础模型的许可才是合同；真正承担风险的，是把它上线的团队。

最强的反方是：切换很便宜。Apache-2.0 的图像模型现成就有，从 FLUX.1-schnell 到本月新发布的几款。用于草稿、样稿或任何“够用就行”的场景，研究许可对你没有成本。

但如果某个模型的画质本身就是产品，这个反方就站不太住：同一榜单上的新许可宽松模型，下载量只有几十到一千左右，说明生态支持还弱（下载量不等于质量）。

如果你在付费项目里用开放图像模型：接入之前，打开基础模型的 LICENSE 文件，而不是看分支的标签；再写下你会切换到的那个许可宽松的模型，用你真实的提示词在它上面跑一遍。这不是法律意见。

如果到 2027 年 1 月 31 日，Qwen 把 2.1 系列改为 Apache-2.0，或者公布可自助签署、条款明确的商业许可，这个判断就不成立了。

Sources来源

WAKENEX / NOTES

Sources来源

READ NEXT接着读

The goods shock is in the pipe, not on shelves.商品价格的冲击已经进了管道，只是还没上货架。

Deleting the expiry date doesn't change the leverage.删掉到期日，杠杆并没有变。

Agents now compete on cheap human steering, not on generation.编码智能体如今比拼的是“让人低成本地掌舵”，因为限制产出的是人的监督注意力，而不是生成能力。