Qwen4-Max Preview Is Live: 14 Days, No Price Yet

Qwen4-Max is finally something you can call — but only just, and only for about two weeks. Alibaba’s Qwen team announced the preview on 11 November 2026, and the flagship tier of the Qwen 4 family is now reachable through a paid Token Plan subscription and, since later the same day, inside Qwen Chat. What arrived is not the Qwen 4 launch. There is no model card, no benchmark score, no per-token price, no downloadable weights and no date for general availability. What arrived is a dated evaluation window, and the window is the story.

Quick answer: Qwen4-Max Preview is live 11–25 November 2026 via Token Plan and Qwen Chat, with a 1M-token context — but no price, benchmarks, weights or launch date yet.

What shipped today

The first Qwen 4 tier anyone can use went live on 11 November 2026 — as a preview with a printed expiry date. The Qwen team’s announcement, posted on X at 02:01 UTC on 11 November, runs three lines: “Qwen4-Max is now in preview on Alibaba’s Token Plan. You can also get early access to the new model on QwenWork, Qoder, Qwen APP, and QwenStudio.” The model ID is qwen4-max-preview. Alibaba’s own documentation puts the window at 10:00 on 11 November to 24:00 on 25 November 2026 (UTC+8), and says the model will be retired when the preview ends — unchanged throughout, with no changes permitted in between.

That framing matters because of what came before it. Alibaba named the Qwen 4 family at its Apsara conference on 22 September 2026 — four tiers, Max, Flash, Plus and a 27B model — and said the family was in training. Its chief technology officer told an Amsterdam stage on 29 October that the family was a week away, according to attendee reports, and the Qwen account posted a teaser on 6 November that did not even name Qwen4. Our earlier Qwen4 preview piece said, correctly at the time, that nothing had shipped yet. That sentence is now out of date in exactly one respect: one tier, for fourteen days, behind a subscription.

What you actually get

The published specification is short enough to quote in full — and the gaps in it are the point. Alibaba’s Model Studio documentation lists the following, and nothing more:

Specification	What Alibaba publishes
Context window	1,000,000 tokens
Maximum output	393,216 tokens — a ceiling the model’s reasoning and its final answer share, not the answer alone
Inputs	Text, image and video
Capabilities listed	Text generation and reasoning, image and video understanding, tool calls, structured outputs
Positioning	Multi-file software engineering, front-end builds, knowledge synthesis, long-running agentic workflows

Two numbers in that table deserve a second look. A 1M-token context puts the preview in the same league as the current flagship class for long-context work, and a 393,216-token output ceiling is unusually large — but because reasoning and answer share that ceiling, a long chain of thought eats the room the answer needs. Neither number says anything about quality: a context window is a capacity, not a capability, and Alibaba has published no benchmark to show how well the model uses it.

Two doors in: Token Plan vs Qwen Chat

There are two ways to reach the same model, and they differ in almost every way that matters. The API route runs through Alibaba Cloud’s Token Plan, a monthly credit subscription, and the consumer route runs through Qwen Chat, where the model appears in the model selector after login.

	Token Plan (API)	Qwen Chat (web/app)
How you get in	Paid Token Plan subscription; dedicated Singapore endpoint (token-plan.ap-southeast-1.maas.aliyuncs.com); an ordinary Model Studio key does not open it	Log in and pick the model — no key, no subscription, no per-token meter visible
Cost	Monthly credit subscription; Personal Edition tiers reported at $6 for 20,000 credits, $10 for 45,000, $18 for 80,000 and $43 for 200,000, plus $15 per 20,000-credit top-ups; a $3-a-month Lite tier is advertised to new users	No per-token charge visible; Alibaba has not published usage limits for the chat route
What you can build on it	Interactive use in compatible coding and agent tools (Cursor, Codex, Claude Code, Qwen Code, Qoder and others); the Personal Edition terms exclude building it into automation scripts or app backends	Chatting and testing only — there is no API behind this door
Expiry	Retired after 24:00 on 25 November 2026 (UTC+8)	Same retirement date

The Token Plan’s restrictions are worth reading literally. Pay-as-you-go is explicitly not supported, the plan operates only in the Singapore region, and its credentials cannot be exchanged with ordinary Model Studio or Coding Plan credentials. This preview is for hands-on evaluation inside approved tools, not for wiring a production dependency to a model that disappears in two weeks. The Qwen Chat door is the only one with no paywall — but it has no published limits either, so heavy use there is on Alibaba’s terms, not yours.

What Alibaba did not ship

Everything a buyer or a builder normally uses to judge a model is missing — and Alibaba has not said when any of it arrives.

No parameter count. The nearest architecture statement from the Qwen organisation is Qwen3.8-Flash-Next, open-sourced in late August 2026, whose model card describes it as an experimental preview of the architecture that will underpin Qwen4 (125 billion total parameters, roughly 6 billion active per step). That describes a preview model, not this flagship.
No benchmark scores — vendor-reported or independent. The Artificial Analysis leaderboard, read on 11 November 2026, carries no Qwen4 entry at all.
No per-token price. The Token Plan sells credits, not tokens, so there is no published dollar figure to compare against rivals.
No weights, no licence, no model card. Qwen3.8-Flash-Next remains the only part of the family anyone can download.
No general-availability date for Qwen4-Max or for the other three tiers — Flash, Plus and 27B — which have not been seen at all.

Early verdicts are already circulating on X, including at least one calling the model a disappointment. Treat every one of them as unconfirmed: there is no benchmark from anyone to confirm or refute them, and a two-week-old preview with no published scores cannot support a verdict in either direction. The honest position today is that nobody outside Alibaba knows how good Qwen4-Max is — including, on the public evidence, most of the people praising or burying it.

Why a 14-day window?

Alibaba is selling an evaluation, not a product — and the expiry date is what makes the evaluation honest. A frozen model with a fixed retirement date means every result gathered against it describes that exact version, and nobody can quietly build a business on a model that will vanish on 25 November. It also lets Alibaba watch real usage — coding, agents, long-context work — before committing to a price, a licence or a benchmark table it would then have to defend. The risk for users is the mirror image: anything you learn about the preview may not transfer to the generally available model, which could be a different checkpoint, a different price, or both.

The comparison that frames the preview best is the model it is meant to replace. Qwen3.8-Max, the current flagship, has been generally available since 3 August 2026 at $2.00 per million input tokens and $6.00 per million output, with the same 1M-token context. Qwen4-Max Preview offers the same context, a much larger output ceiling, and no price at all — a promise with a deadline, where its predecessor is a product with a price list. If you are choosing what to build on this week, that distinction decides for you.

Should you try it?

Try it if you can answer yes to most of these five questions — otherwise wait for the real launch.

Do you already use one of the supported tools? The preview plugs into Cursor, Codex, Claude Code, Qwen Code and Qoder through the Token Plan. If you are not already in one of those, the setup cost buys you two weeks of access.
Is there a job you can finish before 25 November? A real multi-file refactor, a long document synthesis, an agentic workflow you can score yourself. A vague “try it out” burns credits on nothing measurable.
Can you judge it without a benchmark? You will compare it against models you already use, on your own tasks. If you need a leaderboard to decide, this preview cannot give you one.
Are you comfortable with a Singapore-only, credit-based subscription? The Token Plan is a monthly commitment with its own credentials and region. The Qwen Chat door avoids all of that, but gives you no API and no published limits.
Will you treat the results as a snapshot? Whatever you measure describes the preview checkpoint, not the Qwen4-Max that eventually ships. Keep your test prompts and scores — they are only useful if you re-run them on the real thing.

If you answered no to most of those, the sensible move is to wait — and to keep using Qwen3.8-Max, or to try the downloadable Qwen3.8-Flash-Next locally, as our Kimi K3.1 Pro open-weights guide shows for a different open model. The other three Qwen 4 tiers are still unseen, and the generally available flagship still has no date. The next real news will be a benchmark, a price, or a weight file — whichever lands first.

Our take

The window is the product. Alibaba could have launched Qwen4-Max the ordinary way — model card, benchmarks, price, API — and instead it shipped a fourteen-day evaluation with a retirement date and no numbers. That is a defensible way to test a flagship in public without letting anyone build on it, and it is also a way to claim the launch headline while deferring every claim that could be checked. The test that matters now is not whether the preview impresses in a chat window, but whether Alibaba follows it with the three things it withheld: a benchmark, a price, and a date. Until one of those lands, the right response is curiosity, not commitment — try it in Qwen Chat for free, measure it on a real job, and keep your wallet closed until the real model shows up. Desk-researched from Alibaba’s documentation and same-day coverage; not hands-on tested — the preview is Token Plan and Singapore-gated, and no independent benchmark exists yet.

For what else landed this week, see our roundup of 7 new AI tools from 10 November.

Frequently Asked Questions

Is Qwen4-Max out now?

Only as a preview. Qwen4-Max Preview is callable from 11 to 25 November 2026 through Alibaba’s Token Plan and Qwen Chat, then it is retired. The full Qwen 4 family — Max, Flash, Plus and 27B — has no general-availability date, and only the Max preview exists today.

How do I try Qwen4-Max Preview?

The free route is Qwen Chat: log in and select the model. The API route needs a Token Plan subscription in Alibaba’s Singapore region, with its own dedicated endpoint and credentials — an ordinary Model Studio key does not work, and pay-as-you-go is not supported.

How much does Qwen4-Max Preview cost?

Alibaba has published no per-token price. On the Token Plan you buy monthly credits (Personal Edition tiers are reported from $6 for 20,000 credits, with a $3-a-month Lite tier advertised to new users). In Qwen Chat there is no visible per-token charge, but Alibaba has not published usage limits there either.

Can I download Qwen4-Max weights?

No. No weights, licence or model card have been published for any Qwen 4 tier. The only downloadable piece of the family is Qwen3.8-Flash-Next, open-sourced in late August 2026 as an experimental preview of the Qwen4 architecture.

What happens after 25 November 2026?

The preview is retired, per Alibaba’s own documentation. What replaces it — a generally available Qwen4-Max, a price, benchmarks, or a longer preview — has not been announced. Keep any test results you gathered; they describe the preview checkpoint only.

Sources and methodology

Claims in this article rest on Alibaba’s own documentation and the Qwen team’s announcement of 11 November 2026, cross-checked against same-day independent coverage. The Qwen team’s X post (@Alibaba_Qwen, 02:01 UTC, 11 November 2026) is the primary announcement; the specification table (1M context, 393,216-token output ceiling, Token Plan-only access, 10:00 11 November to 24:00 25 November UTC+8 window) comes from Alibaba’s Model Studio documentation as reported in the coverage below. Prices and plan tiers are reported from the Token Plan page, not from a per-token price list, because Alibaba has published none. No benchmark, parameter count or weight file exists to cite, and early verdicts on social media are labelled unconfirmed because no benchmark supports them. Verification date: 11 November 2026. Desk-researched; not hands-on tested.

Sources: OrcaRouter — Qwen4-Max Preview: live 15 days, no benchmark, no price; OrcaRouter — Qwen4-Max Preview in Qwen Chat: a second door, same 14 days; XenoSpectrum — Alibaba’s Qwen4-Max Preview Offers a 1M-Token Context Window, but Access Ends Nov. 25; S3DAT News — Alibaba opens Qwen4-Max preview with a November 25 cutoff; Pasquale Pillitteri — Qwen4-Max arrives in preview on Alibaba’s Token Plan, from $3 a month.

Arva Rangwala

Arva Rangwala covers AI news, AI tools, guides and prompts for OpenAIMaster — what is new, what is worth using, and how to put AI to work.