Image: the Qwen logo — Wikimedia Commons / Mr. Ibrahem, CC0
TL;DR: Alibaba Qwen4-Max preview explained: the first usable Qwen4 model opened on 11 November 2026 as a two-week preview, available only through Alibaba's Token Plan subscription (from $3 for new users) or free in Qwen Chat. It offers a 1M-token context and deep reasoning for coding and agentic work — but there are no published benchmarks, and the preview model is retired on 25 November 2026.
Key Highlights
- Alibaba opened Qwen4-Max-Preview on 11 November 2026 at 10:00 (UTC+8) — the first Qwen4 model anyone can actually use.
- It is a multimodal reasoning model: text, image and video input, deep reasoning, tool calling, web search and structured outputs, built for coding, office work, front-end design and complex agentic tasks.
- Context window is 1 million tokens, with up to 393,216 output tokens shared between the model's thinking and its final answer.
- Access is Token Plan only — no pay-as-you-go API. Token Plan Personal starts at $6 a month ($3 for new users as a launch promo); Qwen Chat offers a free route with no key.
- The preview is frozen and ends at 24:00 on 25 November 2026 (UTC+8) — 21:30 IST — after which Alibaba says the model will be retired. No benchmarks, model card, parameter count or open weights have been published.
What happened?
After almost two months of teasers, Alibaba finally put a Qwen4 model in developers' hands. At 10:00 Beijing time on 11 November 2026, the official Qwen account posted "Qwen4 is coming", and Alibaba listed Qwen4-Max-Preview (model ID qwen4-max-preview) on its Token Plan subscription, with early access in QwenWork, Qoder, the Qwen app and QwenStudio. Later the same day it also appeared in Qwen Chat, Alibaba's free chat surface.
The Qwen4 family was first named on stage at the Apsara Conference in Hangzhou on 22 September 2026, when Qwen project lead Liu Dayiheng said Qwen4 was in training on a new-generation architecture. Four tiers were named — Max, Flash, Plus and a 27B open-weights model — but only Max has shipped, and only as this preview. There is still no date for the full Qwen4 release, or for the other three tiers.
The timing matters because this is a preview with an expiry date, not a launch. Alibaba's own documentation states the window runs from 10:00 on 11 November to 24:00 on 25 November 2026 (UTC+8), that "no changes will be made during this period, and the model will be retired after the preview ends". Anything built on qwen4-max-preview stops working after 25 November — that is 21:30 IST for readers in India.
What Qwen4-Max-Preview can do
Alibaba's Model Studio documentation describes Qwen4-Max-Preview as "the next-generation flagship multimodal model in the Qwen family". The documented capabilities and specs are:
| Capability / spec | What Alibaba documents |
|---|---|
| Context window | 1,000,000 tokens |
| Maximum output | 393,216 tokens, shared by thinking and the final answer |
| Inputs | Text, images and video — screenshot analysis, document understanding, video analysis |
| Reasoning | Thinking enabled by default; reasoning effort low / medium / xhigh (default xhigh) |
| Tools | Function calling, web search, built-in tools, JSON Object / JSON Schema structured outputs |
| API compatibility | OpenAI Chat Completions, OpenAI Responses and Anthropic Messages APIs |
| Designed for | Coding, office productivity, front-end design and complex agentic tasks |
One technical change is worth knowing before you migrate anything: starting with qwen4-max-preview, max_tokens is equivalent to max_completion_tokens — both limit the combined budget for thinking and the answer. Because thinking consumes part of that same budget, a limit previously sized for the answer alone may now truncate output, so leave room for both.
Image: Alibaba Group headquarters, Hangzhou — Wikimedia Commons / Thecraft, CC BY-SA 3.0
How to access Qwen4-Max Preview
There are two doors, and they are not the same. The developer route is Token Plan, Alibaba's monthly credit subscription for coding and agent tools. The steps, per Alibaba's documentation, are: choose Token Plan Personal or Team and get a dedicated API key; use the dedicated Token Plan base URL from the Token Plan guide (the docs list the Beijing endpoint — check the guide for the endpoint matching your region); and set the model ID to qwen4-max-preview. A standard Model Studio or Qwen Cloud API key will not reach it, and pay-as-you-go access is explicitly not supported. Token Plan works with Qwen Code, Claude Code, Cursor, Codex, Cline, OpenCode, Kilo CLI, OpenClaw and Qoder.
The second door is Qwen Chat (chat.qwen.ai), where the same model appeared in the model picker on 11 November. There is no per-token meter, no API key and no subscription to manage — but also no programmatic access, and the same 25 November retirement applies. For most readers who just want to try the model, Qwen Chat is the sensible first stop.
| Token Plan Personal tier | Monthly price | Credits |
|---|---|---|
| Lite | $6 ($3 for new users, launch promo) | 20,000 |
| Essential | $10 | 45,000 |
| Standard | $18 | 80,000 |
| Pro | $43 | 200,000 |
During the preview, Qwen4-Max calls consume credits at 80% off the standard rate, and overnight between 22:00 and 08:00 (UTC+8) consumption drops to 8% of standard. Top-ups cost $15 per 20,000 credits. The new-user Lite price of $3 (19.5 yuan) runs through 30 November 2026. One restriction to note: Token Plan is for interactive use in compatible coding and agent tools only — automated scripts and application backends are outside its terms, so this preview is for evaluating by hand, not for wiring into production.
Alibaba Qwen4-Max preview explained: what Alibaba has not published
The honest headline about this preview is what is missing. Alibaba has published no benchmark scores, no model card, no parameter count, no architecture details, no per-token price and no licence — and no downloadable weights. The Artificial Analysis leaderboard carried no Qwen4 entry at all on 11 November. Alibaba says the model beats Qwen3.8-Max on world knowledge, code generation and sustained performance on long tasks with repeated tool calls, but that is Alibaba's claim, not a verified result, and it should be treated as one until independent benchmarks land.
Early independent testing is already finding quirks: one live test on launch day reported a 300-token reply that contained nothing at all. That is a single test, not a verdict — but it is a reminder that a frozen two-week preview is an evaluation target, not a production dependency.
| Qwen4-Max-Preview | Qwen3.8-Max (current flagship) | |
|---|---|---|
| Status | Preview, 11–25 Nov 2026, then retired | Generally available since 3 August 2026 |
| Context window | 1M tokens | 1M tokens |
| Access | Token Plan only, no pay-as-you-go | Pay-as-you-go API |
| Price | Token Plan credits from $6/month ($3 new-user promo) | $2.00 / $6.00 per million input / output tokens |
| Benchmarks | None published | Public — Artificial Analysis index 45 after reported self-improvement |
| Weights | None | Published as Qwen3.8-2.4T-A95B |
The architecture signal is real even without a model card: Qwen3.8-Flash-Next, released on 28 August, was described by Alibaba itself as an experimental preview of the Qwen4 architecture — an ultra-sparse mixture-of-experts design with reworked attention, 125B parameters plus 51B of N-gram embeddings, activating only about 6B per token, with training cost about one-ninth that of Qwen3.7-Plus. The longer roadmap points further out: models after Qwen4.5 are planned to scale to 5–10 trillion parameters. For now, though, the only Qwen4 fact that matters is the preview window — and it closes in two weeks.
Why it matters for Indian developers
For Indian developers, the practical story is access, not hype. The cheapest way in costs about ₹270 for the first month (the $3 new-user Lite promo), and the free route — Qwen Chat — needs no subscription, no key and no card at all. The 1M-token context is the feature that matters most in daily work: whole repositories, long documents and multi-file front-end builds fit in a single session, which is exactly the agentic coding workload Alibaba is pitching. The overnight 8% credit rate (22:00–08:00 Beijing time, which is 19:30–05:30 IST) also happens to cover Indian evening hours, when many developers do their side-project work.
The catch is the same for everyone: nothing you build on qwen4-max-preview survives 25 November. Treat these two weeks as a free evaluation — compare it against Qwen3.8-Max on your own code, note where it helps and where it fails, and keep your production work on the generally available model. This preview is also a useful contrast to the week's other Chinese flagship story — the Baidu ERNIE 5.5 Flash preview, which arrived with a $0.28 price and the same missing-model-card problem. Both labs are now racing in public, with previews first and proof later.
Should you try it or wait?
✅ Try it now if: you write code for a living and want to judge Alibaba's next flagship yourself — use Qwen Chat for free, or the $3 Lite promo in Token Plan, and test it on a real task before 25 November.
⏳ Wait if: you need benchmarks, a stable API or a model you can build on — the preview is frozen, unbenchmarked and retired on 25 November, and Qwen3.8-Max remains the safe production choice until the full Qwen4 release gets a date.
FAQs
What is Qwen4-Max?
Qwen4-Max is Alibaba's new flagship AI model and the top tier of the Qwen4 family. It is a multimodal reasoning model for coding, office work, front-end design and agentic tasks. As of 11 November 2026 it is available only as Qwen4-Max-Preview, a two-week preview — the full Qwen4 release has no announced date.
How do I access Qwen4-Max Preview?
Subscribe to Alibaba's Token Plan, take a dedicated API key, point your coding tool at the Token Plan base URL and set the model ID to qwen4-max-preview. It works with Qwen Code, Claude Code, Cursor, Codex, Cline, OpenCode and similar tools. The free route is Qwen Chat at chat.qwen.ai, where the model appears in the model picker with no key or subscription.
How much does Qwen4-Max cost?
There is no per-token price, because the preview is sold only through Token Plan credits. Personal tiers cost $6, $10, $18 or $43 a month, with a $3 new-user Lite promo, and Qwen4-Max calls are 80% off during the preview. Qwen Chat access is free. Alibaba has published no pay-as-you-go price for this model.
When does the Qwen4-Max preview end?
The preview runs from 10:00 on 11 November to 24:00 on 25 November 2026 (UTC+8) — 07:30 IST on 11 November to 21:30 IST on 25 November. Alibaba says the model will not change during the window and will be retired when it ends, so any setup using qwen4-max-preview stops working after that.
Takeaway
Qwen4-Max Preview is a genuine milestone — Alibaba's next flagship is finally testable — but it is a 14-day evaluation window, not a launch. Try it free in Qwen Chat before 25 November, judge it on your own work, and wait for benchmarks and a release date before building anything on it. For the wider picture, the week's other big model stories are Thinking Machines' Inkling 2 Small and Google DeepMind's AlphaGenome 2.
Sources
- Alibaba Cloud Model Studio — Qwen4-Max-Preview documentation (primary source: specs, Token Plan-only access, preview window, retirement)
- S3DAT News — Alibaba opens Qwen4-Max preview with a November 25 cutoff
- YottaLabs — Qwen 4: Release Date, Qwen4-Max Preview, and What's Confirmed
- OrcaRouter — Qwen4-Max Preview in Qwen Chat: a second door, same 14 days
- Fello AI — Qwen 4 Release Date: Qwen4-Max Preview Is Out
- ApiDog — What Is Qwen4-Max? Alibaba's New Flagship Model
- AI/TLDR — Qwen4-Max-Preview model page
Comments
Post a Comment