Alibaba Previews Qwen4-Max, Its Next Flagship Model, With Early Access Priced as Low as 10 Percent

Alibaba's Qwen team put a date on the calendar without committing to one: on November 11 the company announced that its next-generation flagship model family, Qwen4, is "about to be released," and in the same breath opened a preview of the top-tier Qwen4-Max to paying subscribers. The preview is available first through Alibaba's Token Plan subscription on the Qianwen AI platform, and Alibaba is pairing the launch with aggressive discounts: Token Plan subscribers get 80 percent off the preview, new users get a further 50 percent off on top, which stacks to early access at 10 percent of the standard price.

The distribution is the more interesting part of the story. Rather than a classic launch event followed by an API opening, Alibaba has pushed the unreleased flagship straight into its own consumer and developer surfaces: Qianwen Office, the Qoder coding tool, the Qianwen mobile app and Qwen Studio all have access to the preview from day one. Developers who want the API can reach it through the Token Plan, a unified-quota subscription that also bundles tool capabilities such as a code interpreter and web search.

On capabilities, Alibaba's description stays at the level of direction rather than evidence. The company says Qwen4 deepens knowledge understanding, builds a more three-dimensional sense of world state, and continues to advance the two capabilities it treats as decisive: coding and agentic task execution. The preview is aimed at complex programming, professional office work, front-end design and operating professional software. Internal tester results, according to Alibaba, showed strong performance in 3D modeling, web design and interactive tasks. The company has not published benchmark scores, parameter counts, context window sizes or comparisons against Qwen3.8, and it has not said whether Qwen4 will be open-weighted.

The pricing ladder around the preview is concrete, though. Token Plan's personal Lite and Essential tiers list at around 39 yuan per month, with new-user promotions bringing that to roughly 19.5 yuan, and the Qwen4 preview discount layered on top during the launch window. Alibaba has not said how long the discount period runs or what Qwen4-Max will cost at list once it reaches general availability.

The sequencing matters for how the open-model market is evolving. Alibaba's previous generation, Qwen3.8, became a workhorse for enterprises that fine-tune open weights, and the Qwen family consistently ranks among the most downloaded open model lines in the world. Opening a preview before the official release, inside a paid subscription and across the company's own apps, turns the launch into a rolling process: real usage and feedback accumulate while the marketing window is still open, and the final release lands on a customer base that has already been onboarded.

It also mirrors a broader shift in how Chinese labs commercialize frontier models. Baidu put ERNIE 5.5 Flash on its Qianfan platform with tiered peak and off-peak pricing before opening weights; Meituan released LongCat-2.5 weights straight to the community; Tencent folded its office agents into a single workbench. The common pattern is to treat a model release not as a single event but as a pipeline from preview to paid usage to open weights, with pricing used to pull demand forward.

For enterprises betting on Qwen, the practical question is what ships at general availability. If Alibaba repeats its Qwen3 playbook — open weights for some tiers, proprietary hosting for the flagship — Qwen4-Max may end up split between a free-to-download sibling and a paid flagship. The company's announcement sets expectations for capability gains without yet providing the numbers to verify them, which makes the eventual release documentation the real test. For now, the cheapest way to judge Qwen4 for yourself costs about two dollars a month.