Alibaba's coding app Qoder has put a model nobody can name into its model picker. Qoduck, described by Qoder itself as a stealth model whose identity has yet to be disclosed, went live on 7 November 2026 with a daily free allowance for every individual user, for a limited time. The launch lands a day after Alibaba's Qwen team teased its next flagship family with a cryptic poster, and users are already guessing the mystery model is Qwen 4 — a guess Alibaba has neither confirmed nor denied.
Free every day, for now
The offer is generous and deliberately simple. Starting 7 November, every new and existing individual user of Qoder International gets a daily free allowance for Qoduck, with no claim required and no paid subscription needed. Free accounts and accounts with a zero Credits balance are eligible too, and paid subscribers get a larger daily allowance than free users.
Within that allowance, calls to Qoduck consume no Credits at all — the rate drops from 0.5 times to zero, even at the highest thinking intensity. Qoder warns responses may slow at peak times because the model is in high demand.
There are limits. Teams and Enterprise subscribers do not see Qoduck in the model picker at all. The promotion applies only to Qoder International, not to the Qoder IDE, the JetBrains plugin, or the Qoder CLI. Qoder has not set an end date, saying only that it will announce one in advance.
A model with no name, no specs, no benchmarks
What Qoder has not published is everything a developer would normally want to know. There is no parameter count, no context window, no benchmark table, and no price beyond the standard 0.5-times Credit rate that applies once the daily allowance runs out.
The event page's own answer to "What is Qoduck?" is a single sentence: it is a stealth model whose identity has yet to be disclosed, and users should try it on their own tasks to see how it performs. More information, it says, will be shared at an appropriate time.
That silence is the point of the exercise, and it is also its limit. A model you can run is more than a poster, but a model with no published capabilities cannot be compared with anything. Until Alibaba names it, every claim about what Qoduck is — or is not — rests on user impressions, not on a model card.
The Qwen 4 question
The timing is what turns a curious promotion into a story. On 6 November, the official Qwen account on X posted a poster of birds flying in the shape of a giant fish, with one line of Chinese and the words "The becoming begins."
The image quotes the opening of a passage from the Zhuangzi, where a fish named Kun turns into a giant bird — a pun on the Chinese pronunciation of Qwen. The post named no model and gave no date. It passed 1.6 million views within 14 hours.
Alibaba has been building toward this moment for months. At its Apsara Conference in Hangzhou on 22 September, the company confirmed Qwen 4 was in training on a new-generation architecture. On 29 October in Amsterdam, a stage slide read "Qwen4 Series Coming Soon" over four names — Max, Flash, Plus and a 27B model.
And in August, the Qwen team open-sourced Qwen3.8-Flash-Next as an early preview of the architecture Qwen 4 will use, with code paths already naming a one-million-token context for qwen4 models. The preview is a real model: 125 billion parameters, with 6 billion active per token, and a context window that stretches to one million tokens.
Against that runway, Qoduck arriving a day after the teaser looks like more than coincidence, and the official Alibaba Cloud account fed the speculation by reposting the announcement with the line "Don't ask where it came from." The repost went out at 09:00 UTC on 7 November, hours after Qoder's own announcement.
But nothing Alibaba or Qoder has published says Qoduck is Qwen 4, and one early user comparison argues it is not even Qwen 4 Flash. The honest status is narrow: a usable model, a hidden identity, and a guess. That is all the evidence there is.
Our Read
Alibaba is running the stealth playbook that has become familiar in the open-model world: put an unnamed model in front of real users and let the community do the identifying. The approach has a precedent with a warning attached. It is a cautionary tale.
A mystery model on OpenRouter earlier this year was widely guessed to be Qwen 4 before it was revealed as Z.ai's GLM-5.3-Flash. Guesses are cheap; model cards are not. Qwen's own Qwen3.8-Flash-Next preview shows how Alibaba usually handles this: publish the architecture early, with weights and a model card, and let the community examine it.
There is also a practical logic to the free allowance. Qoder gets real coding workloads against an unidentified model at no cost to users, and the daily reset keeps the trial going without committing to a price. That is a cheap way to test a model at scale.
If Qoduck is Qwen 4, this is the softest of soft launches — a flagship smuggled into a coding app a day after its teaser. If it is something else, it is still a new Alibaba coding model, free for individuals.
Coding assistants are the sharpest edge of the model race, and either way the reveal matters more than the teaser. The reveal has not happened yet. What to watch is simple: a model card, a Hugging Face repository, or a benchmark table with Qoduck's name on it.
Why It Matters
For developers, the immediate value is concrete: a free daily allowance for a model from one of China's largest AI labs, usable today in Qoder International, with no subscription and no Credits burned. It costs nothing to try, and the daily allowance resets every morning.
For the wider industry, Qoduck is the latest signal that Alibaba's flagship launch is measured in days or weeks, not months. And for anyone tracking the open-model race, it is a reminder that the next frontier model may arrive first as a nameless entry in a picker, not on a keynote stage.
Background
Qoder is Alibaba's agentic coding platform, and Qwen is the model family behind it. The current flagship, Qwen3.8-Max, was released in August 2026, with open weights following days later. Qwen models power Apple Intelligence in China and ship inside coding tools from other vendors.
Source: Qoder's Qoduck event page and CellCog's Qwen 4 tracker
0 Comments