Alibaba's official account posts teaser image quoting the "Zhuangzi," without naming "Qwen 4"; release timing not announced

Official account posts "The becoming begins."

The official X account of "Qwen," the AI model of China's Alibaba Group, on Nov. 6 (the evening of the same day Japan time) posted the line "The becoming begins." with an image. The image shows the Qwen logo, a huge flock of birds flying over the sea at dusk and the text "北冥有鱼 其名为 _." The post gives no model name, release date or description of performance. Alibaba has not said what it is teasing, and the view that it is a teaser for the next-generation model "Qwen 4" remains an interpretation by users and media outlets.

"北冥有鱼 其名为" is the opening passage of the Chinese classic the "Zhuangzi." The original continues, "In the northern sea there is a fish, and its name is Kun," and the giant fish Kun eventually transforms into the bird "Peng." The image leaves the name blank. Chinese IT outlet MyDrivers reported on Nov. 6 that the blank is filled by "鲲" (Kun), noting speculation among internet users that it sounds similar to "Qwen," and called this theory plausible. Alibaba has not explained this reading.

The developer account @QwenDevs is also said to have quoted the post, writing "Qwhen? Qsoon!" and explaining the "Zhuangzi" passage in English. ORIENS could not open the original of this quote post, and its wording and posting time have not been confirmed.

"Qwen4 next week" is an attendee's report, not in company documents

Behind the post being taken as a Qwen 4 teaser is a remark made on Oct. 29. At "Apsara Conference," an event Alibaba Cloud held in Amsterdam, the Netherlands, an attendee posted on X that the company's chief technology officer (CTO), Feifei Li, said "Qwen4 next week." On the same day, Rui Ma, who comments on China's tech industry, also relayed the remark, citing someone who was at the venue, and pointed out that the slide still read "coming soon."

Taken literally, "next week" means Nov. 2 to 8, and the post came out on the Friday of that week. However, Alibaba's documents mentioning Qwen 4 go only as far as the Sept. 22 statement that it is "currently in training." The company describes "Qwen3.8-Flash-Next," released in August, as a preview of the architecture that will underpin Qwen 4, but has not given a schedule for Qwen 4 itself. Regarding Qwen 4's release timing, information pointing to "the 24th" and "within October" also emerged in October, and both proved wrong.

Product names, size and release order all unannounced

The four variants circulating as Qwen 4 product names, Max, Plus, Flash and 27B, are what attendees reported seeing on slides in the Oct. 29 presentation. TestingCatalog, which tracks new AI products, responded to the post by predicting that these four would be released soon. MyDrivers expects the four not to be released at once but in sequence from November to December, and also mentioned the possibility that the largest would exceed 3 trillion parameters. These are all outlets' projections, and Alibaba has not announced product names, sizes or the order.

In October, settings were added early to the development tool "Qwen Code" for model names starting with qwen4. These settings work by matching names and do not mean the specifications are finalized. As of early Nov. 7 (Japan time), the latest post on the official X account was this teaser, and there was no model bearing the Qwen 4 name in Qwen's repository on Hugging Face. Pricing, performance figures and the license have not been announced either.

How much of the weights will be released is the focus

What developers are watching is the scope of release of the weights, the core data of the model. In the previous generation, the 27B "Qwen3.8-27B" was released under the permissive Apache 2.0. "Qwen3.8-Flash-Next" was released under the proprietary "qwen-community-1.0" license, and "Qwen3.8-2.4T-A95B," with 2.4 trillion total parameters, under the proprietary "qwen3.8-max" license. Qwen3.8-Max is described as the official version based on this 2.4T model with added features such as image input. Whether Qwen 4 will come with the same terms is unknown.

Qwen's development structure changed this year. Lin Junyang, its technical lead, left the company in March, and in September Liu Dayiheng became head of large language models. Startup Fortune reported that developers have raised questions about the company's stance on releasing weights.

Rival Kimi K3.1 Pro raises the bar for open models

Among Chinese companies, Moonshot AI released the weights of Kimi K3.1 Pro on Nov. 4 (Nov. 5 Japan time). It scored 46 on U.S.-based Artificial Analysis' overall index (version 5), above the 32 of DeepSeek and Zhipu, previously the highest among models with released weights. It falls short of the 58 of top-ranked Claude Opus 5.5. If Qwen 4's weights are released, this 46 will be the benchmark for comparison.

Returns on AI investment are the earnings focus

Whether new models translate into cloud revenue will be the focus of the next earnings report. Alibaba's capital expenditure in the April-June 2026 quarter rose 75% to 67.7 billion yuan from 38.7 billion yuan a year earlier. Revenue from its AI cloud and computing services unit rose 45% to 48.4 billion yuan from 33.4 billion yuan a year earlier. The date of the July-September earnings release had not been found as of early Nov. 7; the previous year, it reported on Nov. 25.