Alibaba Launches Qwen4-Max Preview: 1M Context, Access & What Is Missing
Alibaba has opened a two-week preview of Qwen4-Max, its next-generation flagship AI model with a 1M-token context window, but no benchmarks or price yet.
Image: Wikimedia Commons / Thomas LOMBARD, designed by HASSELL (architects)
HANGZHOU, Nov. 13 — Alibaba has opened a preview of Qwen4-Max, the flagship model of its new Qwen 4 family, giving developers a first look at its next-generation AI system with a 1-million-token context window.
The preview opened on November 11 and runs until November 25, after which Alibaba says the model will be retired. It is available only through the Token Plan subscription and Alibaba's own apps, with no published benchmarks, model card or per-token price.
Key Takeaways
- Qwen4-Max Preview opened Nov. 11, the first model of Alibaba's Qwen 4 family, a multimodal reasoning model with a 1M-token context window.
- Access is limited to the Token Plan and Alibaba's own apps: Qwen Studio, the Qwen app, QwenWork and Qoder.
- The preview ends Nov. 25 and the model will be retired; Alibaba has published no benchmarks, weights or price.
- Three more tiers — Flash, Plus and a 27B model — have been named but have no release date.
What Is Qwen4-Max and What Can It Do?
Qwen4-Max is the top tier of the Qwen 4 family, Alibaba's next-generation flagship multimodal model. It is built for reasoning, coding, design, office work and agentic tasks, with text and image input. The 1-million-token context window is the one hard specification published: it lets the model hold far more text in a single conversation than most models.
The rest of the family was shown at the Apsara Conference in Hangzhou in September, when Alibaba said Qwen 4 was still in training. The four tiers are Max, Flash, Plus and a 27-billion-parameter model for local use. Only Max has shipped, and only as a preview. Rivals are making their own big moves in the same week.
What is missing matters as much as what is here. There is no parameter count, no architecture detail, no model card, no licence and no benchmark score, so the model cannot yet be measured against its rivals. Rival models from OpenAI, Anthropic and Google all carry published scores; Qwen4-Max carries none.
How to Try Qwen4-Max Preview Today
The model is available in five places: Qwen Studio, the Qwen app, QwenWork, Qoder, and the API on Alibaba Cloud's Token Plan. In Qwen Studio it appears in the model picker as Qwen4-Max-Preview. The API model ID is qwen4-max-preview.
The Token Plan is a monthly subscription billed in credits, not pay-as-you-go, so there is no published price per million tokens. No changes will be made during the two weeks, and the model retires when the preview ends on November 25. The model is aimed at agentic tasks, a category drawing close scrutiny.
There is no open version: Alibaba has published no weights, and no Qwen 4 repository has appeared where it normally publishes its open models.
When Will the Full Qwen 4 Family Be Released?
Alibaba has given no date for the full Qwen4-Max release or the other three tiers. The announcement says only that Qwen4 is coming and describes a preview, not a finished product. Flash, Plus and the 27-billion-parameter model have no dates either.
The preview is a test, not a launch. Developers are asked to try it on a real problem; the two-week window is the only period guaranteed. What comes after, what it will cost, and whether it will behave the same have not been announced.
For now, Qwen4-Max is a name, a model ID and a two-week trial. The numbers that decide whether a model matters — benchmarks, price and open weights — are still missing. The gap between announcements and measurable products is widening across the AI industry.
| What | Detail |
|---|---|
| Model | Qwen4-Max Preview, Qwen 4 flagship |
| Status | Preview; Nov. 11-25, then retired |
| Context window | 1 million tokens |
| Access | Token Plan; Qwen Studio, app, QwenWork, Qoder |
| Price | No per-token price; credits billing |
| Benchmarks | None published |
| Full release | No date; Flash, Plus, 27B also undated |
Frequently Asked Questions
Is Qwen4-Max free to use?
No. The preview runs only inside the Token Plan, a paid subscription billed in credits, with no free trial. Qwen Studio chat needs a login.
What is the Qwen4-Max context window?
One million tokens — the one hard specification Alibaba has published for the model.
Does Qwen4-Max have benchmarks?
No. Alibaba has published no benchmark scores, model card, parameter count or architecture detail, so the preview cannot yet be compared with rivals.
When does the Qwen4-Max preview end?
November 25, 2026. Alibaba says the model will be retired after the preview ends, with no changes during the two-week window.
