Alibaba has opened a preview of Qwen4-Max on its Token Plan1, the company’s official Qwen account said on X.

The post adds a new Qwen-Max model to Alibaba’s lineup, but it gives availability details rather than testing data. A retrieved excerpt from Bloomberg Technology2 said Alibaba claimed performance that rivals Anthropic and OpenAI; Alibaba’s announcement includes no benchmark table, and AI/TLDR’s summary of Qwen Cloud documentation3 said Alibaba had published no model card or per-token pricing at launch.

Background on the line comes mainly from that AI/TLDR page3, which lists Qwen2.5-Max from January 2025, Qwen3.7-Max from May 2026, Qwen3.8-Max from August 2026 and Qwen3.8-Max-0902 from September 2026 before Qwen4-Max-Preview. The same page said Qwen4-Max-Preview runs only through the Token Plan subscription, not pay-as-you-go API access, from 10:00 on 11 November 2026 to 24:00 on 25 November 2026, UTC+8, with the model frozen during that window and retired when it ends.

In its X post, Alibaba described Qwen4 as offering “deeper knowledge,” “richer world understanding,” and “stronger coding and agentic capabilities,” and as built for real-world tasks1 including complex coding, knowledge work, front-end builds and professional software workflows. Alibaba also said early access is available on QwenWork, Qoder, Qwen APP and QwenStudio.

Technical details in the retrieved material come from secondary reports. AI/TLDR3, summarizing Qwen Cloud documentation, said Qwen4-Max-Preview accepts text, images and video, has a 1-million-token context window and supports up to 393,216 output tokens. The page also said thinking is on by default and controlled with a reasoning_effort setting.

Parameter reporting is inconsistent. TechFlow4, citing Jin10 Data, reported that Qwen4-Max has 2.4T total parameters, native visual capabilities and a million-token context. Separately, TechFlow cited Qwen officials for performance claims that the model ranked sixth globally in the frontend coding blind test CodeArena and placed in the top tier on Agents’ Last Exam and Vals Index. Those are vendor claims and the cited materials include no independent test results. AI/TLDR3 said Alibaba published no parameter count at launch and lists 2.4T weights for the earlier Qwen3.8-Max, not Qwen4-Max.

The retrieved X post carries no visible timestamp; secondary reports place the rollout on or around November 16, while AI/TLDR3 said the preview window began on November 11, UTC+8.