At 23:38 on November 7, 2026, Qoder, the coding tool owned by Alibaba, published a short post on its official WeChat account. The page header reads "Qoder, November 7, 2026, 23:38, Zhejiang", and the title translates roughly as: "Twenty-four hours after launch, the whole internet is guessing who TA is; we only want to ask: how much work can it do for you?" The subject is a black duck: an anonymously launched model that Qoder calls Qoduck. No launch event, no parameter sheet, no benchmark poster. The official wording is blunt about what is missing: "no official announcement of its origin, and no benchmark posters." As of November 9, 2026, the guessing game about the model's identity is still running, and Qoder has added nothing about where it comes from. This article does not guess. It takes apart two things instead: what the official post actually says and what it deliberately leaves out, and what a user can realistically get inside the limited-time free window.
A quick note on how this piece fits with the three companions shipped in the same batch. This is the hotspot column, and it covers only the Qoduck anonymous launch and the boundary of the official wording. If you care about cluster-side inference infrastructure, our open-source column takes apart Kinference, a routing control plane for Kubernetes and vLLM, in the Kinference deep dive. If you want to compare hosted model APIs on what their free tiers actually give you, our comparison column lines up five providers in the free-tier comparison. If you need to migrate off a platform that is shutting down, our SOP column walks through the damage control step by step in the Yuanqi migration SOP. Four articles, four lanes, no overlap.
A launch without a launch event
By industry convention, a new model makes its entrance with a standard three-piece set: a launch event, a benchmark poster, and a parameter sheet. Qoduck's entrance skipped all three. The official post reads, in translation: "Qoder quietly launched an anonymous model: Qoduck." "It has no launch event, no parameter sheet, not even a name — only a black duck." "No official announcement of its origin, and no benchmark posters." Three sentences, and the shape of the launch is fully described: the model lives inside Qoder, its product name is Qoduck, its public face is a black duck, and everything else is off the table.
Pin the timeline first. The launch happened at 23:38 on November 7, 2026, per the page header of the official WeChat post. As of November 9, 2026, various technology aggregators have already added Qoduck to their lists of newly released models. Those compilation dates are aggregation dates, not the launch date, and the two are roughly two days apart; citing them interchangeably is a small but real factual error that has already started to spread.
Then look at what "anonymous" concretely hides. On the naming layer, Qoduck is a product name; the official post gives no model family, no version number, and no background of any kind. On the origin layer, it does not say who trained the model or how it relates to any existing model. On the technical layer, the parameters and the context length — the two specifications that usually head every release note — are absent. In other words, the sum of confirmed information is this: an anonymous model called Qoduck exists inside Qoder, its mascot is a black duck, and the rest is blank.
That blank is exactly why this article sets one rule up front: what the official post does not say, we do not guess. The various identity theories circulating in the community are out of scope, and the guess lists that some aggregators have compiled are not treated as clues here. This is not written out of caution for its own sake. It is the only defensible way to cover this launch: the official wording is finite, and every sentence added beyond it turns reporting into fabrication.
The black duck deserves a note of its own. Model reveals in coding tools usually chase a high-tech aesthetic, and a duck runs the opposite way, which is likely why it sticks in memory. Software engineers also have a long-standing tradition of rubber-duck debugging — explaining your code aloud to an inanimate duck until the bug surfaces by itself. Whether the black duck is a nod to that tradition is something the official post does not say, and we will not interpret it on Qoder's behalf. What is confirmed is simpler: the image appears in the official post, and it is the only visual symbol this launch was given.
What Qoder said, and what it did not
Splitting the official wording into two columns is where this piece earns its keep.
What the official post does state falls into three categories. The first is the launch fact itself: Qoder quietly launched an anonymous model named Qoduck, presented as a black duck, at 23:38 on November 7, 2026. The second is the free-access policy: during the event period, Qoduck is limited-time free for Personal plan users, with a defined way of handling overflow usage, detailed in the next section. The third is the runtime boundary: demand has been high enough that short queues may form at peak hours, and the CLI and IDE entry points are not yet open.
What the official post does not state is a longer list: the model's origin and trainer, its parameter count, its context length, the specific numbers behind the free quota, the end date of the limited-time free period, and any form of benchmark data. Filling any of these six gaps with a figure or a claim would be fabrication. One warning is worth repeating: quota numbers and deadline dates circulating on secondary aggregation pages do not appear anywhere in the official post. When you meet such a claim, check it against the source post before passing it on.
There is also a layer of attribution that is easy to blur. Qoder is a coding tool owned by Alibaba, and that corporate relationship is explicit. But who made the Qoduck model itself, and whether it has any connection to Alibaba's own model lineup, is something the official post never addresses. Tool ownership and model ownership are two different questions, and one cannot be substituted for the other. As of November 9, 2026, that information gap remains exactly where it was on launch night.
Expect the surrounding noise to grow before it settles. With an anonymously launched model, secondhand retellings always outrun the source post: retellings add details, details drift into new wordings, and wordings harden into conclusions. A reader does not need to chase every retelling. Two habits are enough: treat the official WeChat post as the only reference for verification, and keep the six-gap list above in mind — whenever a concrete number or date appears for any of those six gaps, treat it as retelling noise until the official channel confirms it.
Where the capability claims come from
Every capability statement Qoder makes about Qoduck traces back to a single sentence: "Based on comprehensive community reviews and internal testing by Qoder ambassadors, Qoduck has several standout strengths" (translated). That sentence governs how every claim in this section must be cited. These are officially relayed community feedback and ambassador impressions, not official benchmarks. There are no scores to quote, and there is no scoreboard to rank Qoduck on.
Four strengths are relayed in the official post. First, speed. The official relay describes it as "the only unanimous conclusion across all reviews" — the one point on which every community reviewer and ambassador tester agreed. Second, steadier performance on long-horizon tasks. The relayed phrasing is a higher probability of completing a task in one pass, meaning fewer mid-course failures on multi-step work. Third, sound reasoning. The relay mentions small reasoning puzzles such as the "candy question" being answered correctly, and community commentary quoted in the post includes the phrase "it has a brain". Fourth, complete delivery. The relayed cases include a Trello board task finished with "zero defects", a 30-second promotional video, and the community verdict "basically first-tier level, and the key is that it is free right now".
How you read these four claims matters. Speed, steadiness, reasoning, and delivery are all qualitative descriptions with no quantitative framing attached: no number for how fast, no baseline against which the one-pass completion rate was measured, and "first-tier level" is a community member's phrasing rather than an evaluation result. Read them as a signal that the model is worth trying with your own hands, not as performance conclusions to quote. The only reliable way to judge whether Qoduck can carry your workload is to run a slice of your real work through it while the free window lasts.
The division of labor with our capability comparisons belongs here too: this article judges no model against another. The systematic comparison of terminal agents and their models lives in our Claude Code versus Cursor versus Codex review, and the two pieces complement each other.
Inside the limited-time free window
The official policy reads, in translation: "During the event period, Qoduck is limited-time free for Personal plan users, and the 'Ultra-high' thinking intensity does not deduct Credits." And: "Once you exceed the free quota, you can either come back the next day, or spend credits to keep using it." Combined with the runtime boundaries, the practical situation can be laid out in five points.
First, who gets it free. During the event period, the offer covers Personal plan users. Whether team plans are included is not addressed in the official post. Second, what is free. Thinking at the "Ultra-high" intensity does not deduct Credits, so running Qoduck at that setting does not consume your credit balance. Third, what happens on overflow. Two official paths: wait for the next day, or spend credits to continue immediately — the choice is the user's. Fourth, until when. No end date for the event is given anywhere in the official post, and as of November 9, 2026 no supplementary announcement has appeared, so every concrete "ends on" date in circulation is unsupported. Fifth, where. The CLI and IDE entry points are not yet open; the model currently lives inside Qoder's existing product. And because demand is high, short queues may form at peak hours, with the official advice being to try at off-peak times.
One piece of strategy beyond the five points: since the deadline is undefined, treat the free window as an open-ended trial rather than a guaranteed long runway. Try early, verify early, and do not plan around the assumption that the window will still be there next month. The official closing line is worth quoting here as well, in translation: "We believe the best model is in every real workflow." That sentence is consistent with the whole anonymity play: the way to evaluate a model is not to read its poster but to watch it handle your own work.
Whom the anonymity hands judgment to
Finally, the interesting part of the story. Anonymous launches are not new, but Qoduck carried the anonymity to an unusual degree: no origin, no parameters, no benchmarks, and not even one line of official self-assessment. Every capability claim in the post is voiced through community reviews and ambassador testing. The objective effect of that staging is to push the right of judgment off the launch-event stage and into real workflows — and the official line "we believe the best model is in every real workflow" reads as the footnote to exactly that narrative.
For practitioners, two lessons stand out. The first is information discipline. Faced with a model that has no origin, no parameters, and no scores, the trustworthy information boundary is the official wording itself. In multi-layer retelling, "community feedback" gets polished into "official conclusions" with remarkable speed; citing well means keeping the provenance attached, the way this article has tried to do. The second is verification method. Rather than watching the identity guessing game from the sidelines, spend the free window throwing your daily tasks at the model. Speed, long-horizon steadiness, and delivery quality — the very traits relayed in the official post — are precisely the dimensions you can test against your own workflow without any scoreboard.
As for the black duck itself, the official post explains neither the image nor the name, and this article offers no interpretation of either. A black duck with no name, no parameters, and no benchmarks, two days into a launch, with the internet still guessing who it is — that guessing game is itself the core of this launch's distribution design. As of November 9, 2026, the confirmable facts remain what they were on night one: Qoduck is an anonymous model inside Qoder, limited-time free for Personal plan users, origin unknown.
Related reading Kinference: an inference routing control plane for Kubernetes and vLLM — the same batch's open-source deep dive. Free hosted LLM API tiers compared — what five providers' free tiers give, what they limit, and what they take in exchange. Yuanqi shutdown migration SOP — step-by-step damage control for a platform shutdown. Five GitHub AI tools, metadata only — objective parameter sheets, a neat contrast to a launch with none. Choosing an agent concurrency runtime — suspended, isolated, or resident.