Alibaba’s Qwen team posted an image and the line “The becoming begins” on its official X account on Nov. 6, a teaser that kingy.ai called an interpretation, not a product announcement, of the coming Qwen4 artificial intelligence models.
The image shows, in Chinese, the opening line of the first chapter of the Zhuangzi, a Daoist classic, which reads roughly “In the Northern Ocean there is a fish, named,” with the fish’s name left blank. Neither the text nor the image mentions Qwen4. The link rests on earlier Alibaba statements and an event remark.
Zhuangzi was a Daoist thinker of China’s Warring States era in the fourth century BC, and the book that bears his name is a collection of parables. It opens with a fish called Kun, said to be so large that no one knows how many thousand li it measures. The fish turns into a bird, Peng, whose wings hang like clouds and which must climb 90,000 li, a traditional Chinese distance of about half a kilometre, before the wind can carry it south. A cicada and a dove laugh at the flight.
Readers have long taken the story as a lesson in scale, perspective and transformation. Digitalmarketer read the teaser as a sign the current generation is giving way to a bigger successor, and a briefing on kimkj.com, a personal AI-news site, reads the blank as Kun, a pun on “Qwen.” Alibaba has not said what the image means, according to the coverage reviewed.
The Qwen4 lineup, as reported, has four tiers: Max, Plus, Flash and a 27-billion-parameter model that can run on a personal computer. Latenode reported the names on Oct. 29, attributing them to Alibaba’s own event recap. The kimkj.com briefing expects the four to be unveiled one after another in November and December but calls that timing unconfirmed. Parameters are the adjustable values that make up an AI model.
Is it for a PC? Only the 27B tier, possibly, and only according to the reports. The kimkj.com briefing says users can download it and run it on a personal computer, and that its predecessor, Qwen3.8 27B, was a benchmark for models run locally on PCs. Alibaba hasn’t published hardware requirements or confirmed that the 27B model will be downloadable. Max, Plus and Flash would be cloud models. Max will probably be far too large for a PC, since the current Max has 2.4 trillion parameters.
As a rough rule of thumb (not an Alibaba figure), a 27-billion-parameter model compressed to 4-bit precision needs about 15 GB of memory. That fits a high-end gaming graphics card with 16 to 24 GB, or a Mac with 32 GB or more. At full precision it needs about 54 GB, which is workstation territory.
Alibaba has published no release date, prices, benchmarks or downloadable model files. Its most recent formal statement, a Sept. 22 press release tied to its Apsara Conference, said Qwen4 “is currently in training” and projected later Qwen4.5 and Qwen5 models of five to 10 trillion parameters. An attendee at the company’s first international Apsara Conference, in Amsterdam on Oct. 29, said chief technology officer Feifei Li told the audience Qwen4 was coming “next week,” FelloAI reported. That week, Nov. 2 to 8, has ended, and no release had been reported as of Nov. 6.
Alibaba’s largest current model, Qwen3.8-Max, has 2.4 trillion parameters, about 95 billion of them active at a time, slightly fewer than Moonshot AI’s 2.8-trillion-parameter Kimi K3. Observers speculate Qwen4 could pass three trillion, which Alibaba has not confirmed. No Qwen4 repository existed on Hugging Face, the model-sharing site, as of Nov. 5, FelloAI reported, and Alibaba has not said whether any Qwen4 weights will be made public.
Source: kingy.ai; Digitalmarketer; kimkj.com; Facts and Details; Alibaba Cloud; FelloAI; Latenode; The Next Web
