AI-China market & macro brief: Qwen's open weights stay wide, but the paid route is moving into Model Studio and Coding Plan

A real photograph of Alibaba's Hangzhou headquarters fits this article because the story is not one isolated model launch. It is about how Alibaba is arranging the route from open Qwen releases to managed developer traffic inside its own access stack.[5]

As of 2026-04-29 UTC, the sharper way to read Qwen is to separate two things that are still being discussed as if they were one. On one side, Alibaba keeps the open-model story wide: the official Qwen3.6 repository says Qwen3.6-35B-A3B and then Qwen3.6-27B were released to Hugging Face and ModelScope, while Qwen Studio, Model Studio API, and Qwen Code remain official access paths.[3] On the other side, the commercial route is being narrowed and reorganized. In April, Qwen Code first offered Qwen 3.6 Plus free inside the product with 1,000 calls per day and 1 million-token context, then days later announced that the Qwen OAuth free tier was discontinued and pointed users toward Alibaba Cloud Bailian / Model Studio and other paid providers instead.[1][2] Put beside Alibaba Cloud's own Coding Plan documentation, the market signal becomes clearer: the open frontier is still broad, but the monetizable traffic lane is being pulled into a managed routing stack.[1][2][3][4]

That distinction matters in ai-china because Chinese model competition is no longer only a release race. The harder question is who controls the path from a public model announcement to repeated, paid developer use. Qwen's April documentation suggests Alibaba wants both layers at once. It still wants the reach that comes from open weights and broad compatibility, but it also wants serious coding usage to settle inside product surfaces, account systems, and quota regimes it directly manages.[1][2][3][4]

Image context: the cover uses a real Wikimedia Commons photograph of Alibaba's Hangzhou headquarters. That is the right visual here because the article is about company-level routing strategy. The key issue is not what one benchmark chart looks like by itself, but how Alibaba moves developers from public model visibility into controlled usage channels.[5]

The open side is still intentionally wide

It would be a mistake to describe Qwen's April posture as retreat from openness. The official Qwen3.6 repository says the 35B-A3B model became available on 2026-04-16, and the 27B dense model followed on 2026-04-22, with official distribution through Hugging Face and ModelScope.[3] The same repository also describes Qwen Studio as the first-party place to try the models, Alibaba Cloud Model Studio as the official API path, and Qwen Code as the terminal agent optimized for Qwen models.[3]

That combination matters because it keeps Qwen visible across several developer strata at once. Self-hosters can still take weights from open-model channels.[3] Builders who want hosted access can use Model Studio's API layer.[3] Developers who want a branded coding workflow can stay inside Qwen Code.[1][3] Alibaba is not closing doors. It is multiplying them.

The important market point is that these doors no longer mean the same thing. An open release on Hugging Face or ModelScope preserves adoption breadth and mindshare, but it does not automatically tell you where the long-lived revenue or product lock-in will sit. Alibaba's own docs suggest that the value is increasingly being built one layer above the release itself, in the place where keys, quotas, workflow defaults, and usage habits are decided.[1][3][4]

April turned "free access" into a routing decision

The April 8 Qwen Code weekly update is revealing because it still presents generosity as part of the product pitch. Qwen says Qwen 3.6 Plus is officially available in Qwen Code, free to use, with 1,000 calls per day and a 1 million-token context window, and it places Alibaba Cloud Coding Plan next to that announcement on the same update page.[1] That is not just a model note. It is a funnel design. Users are invited to try the strongest first-party coding experience inside Qwen's own surface before they have to think hard about external integration.[1]

Then the tone changes. The April 16 Qwen Code update says that, starting 2026-04-15, the Qwen OAuth free tier has been discontinued, though existing users can consume their remaining quota.[2] The same notice recommends migration to Alibaba Cloud Bailian first, then to OpenRouter and Fireworks.[2] That ordering matters. Qwen is not only ending one free path. It is telling users which replacement lane it would most like them to choose.

This is why the free tier should be read as a routing layer rather than a permanent economic promise. Free usage inside the branded product can accelerate habit formation.[1] But once users need continuity, scale, or predictable integration, the docs point them toward managed channels that Alibaba or its preferred stack can meter and support more directly.[2][4]

Coding Plan shows where controlled usage becomes the product

Alibaba Cloud's Coding Plan documentation makes the monetization logic unusually explicit. The page says the plan is a $50/month subscription for AI coding tools, that the Lite plan stopped accepting new subscriptions on March 20, 2026, and that the Pro plan carries quota windows of 6,000 requests per 5 hours, 45,000 per week, and 90,000 per month.[4] It also says each query consumes quota differently depending on task complexity, context length, and tool usage.[4]

That already looks less like raw model resale and more like a managed workflow contract. The plan is not marketed as generic token access. It is framed around coding tools, request windows, and predictable monthly usage.[4] The documentation sharpens the boundary further by saying Coding Plan has its own API keys, formatted as sk-sp-xxxxx, and its own base URLs, which are different from the general pay-as-you-go Model Studio API and must not be used interchangeably.[4]

That separation is strategically important. It means Alibaba is not merely charging for compute. It is sorting developer traffic into different commercial lanes with different credentials, allowed behaviors, and support assumptions.[4] The same page even warns that Coding Plan is for interactive coding tools rather than automated scripts or custom application backends, and that misuse can lead to suspension or revocation.[4] In other words, the product is not only selling a model. It is defining the approved shape of use.

The macro signal is a split between reach and control

Taken together, these pages point to a broader AI-China pattern. Chinese model companies still need the reach benefits of open release culture, broad compatibility, and visible benchmark progress.[3] But the more valuable layer is increasingly the one that decides where repeated developer behavior settles: inside a first-party coding client, a pay-as-you-go API account, or a subscription plan with provider-defined keys and workflow rules.[1][2][4]

Alibaba's April Qwen stack is a clean example of that split. The company keeps open weights and official community channels active enough to preserve ecosystem relevance.[3] At the same time, it is tightening the economic handoff from experimentation to payment: first-party free usage inside Qwen Code for discovery, then Model Studio or Coding Plan for sustained work, with credential boundaries and quota logic making the route legible.[1][2][4]

That is why this matters as a market-and-macro brief rather than as one more release note. The underlying contest is moving from "who has a strong coding model" to "who owns the route from open model attention to paid workflow dependence." Qwen's April documentation suggests Alibaba understands that the answer will not be decided by openness alone or by paywalling alone. It will be decided by how effectively the company lets both coexist while steering the serious traffic into its own managed lanes.[1][2][3][4]

cronfeed.work

AI-China market & macro brief: Qwen's open weights stay wide, but the paid route is moving into Model Studio and Coding Plan

The open side is still intentionally wide

April turned "free access" into a routing decision

Coding Plan shows where controlled usage becomes the product

The macro signal is a split between reach and control

Sources

Recommended In ai china