Qwen4: Alibaba’s new model family, Qwen4-Max price, access and comparison
Everything about Qwen4 in one place: Qwen4-Max Preview facts, where to use it, Token Plan pricing and a table against Claude Fable 5.2, GPT-6 and Gemini 4 Carbon.
👁 111 reads
Qwen4 is the fourth generation of Alibaba’s language models. The first member of the family, Qwen4-Max in preview, arrived on 11 November 2026: in the Qwen Studio chat, in the Qwen app, in QwenWork and Qoder, and in the API through the Token Plan subscription. This guide collects what is known for certain about Qwen4, where you can use it, what it costs and how it compares with Claude Fable 5.2, GPT-6 and Gemini 4 Carbon. I update it with every new model in the family.
As of 11 November 2026. Available model: Qwen4-Max-Preview (qwen4-max-preview). Preview until 25 November, 16:00 UTC. Context 1 million tokens. Vendor benchmarks: none. OpenRouter: not listed. My test: not yet, because the model is not in the API without an Alibaba subscription.
What is known about Qwen4-Max
| Feature | Qwen4-Max-Preview | Qwen3.8 Max (predecessor) |
|---|---|---|
| Release | 11 Nov 2026 (preview) | 3 Aug 2026, snapshot 0902 from 2 Sep |
| Context | 1,000,000 tokens | 1,000,000 tokens |
| Input | text, image, video (in chat also documents and audio) | text, image, video |
| API access | Token Plan only (subscription) | pay per token: Alibaba Cloud, OpenRouter |
| Price per million tokens | not published | $2 input, $6 output |
| Artificial Analysis index | not listed | 45 (27 Sep 2026) |
Sources: Alibaba Cloud Model Studio documentation, Qwen Studio model list, OpenRouter, Artificial Analysis. As of 11 November 2026.
Alibaba describes Qwen4-Max as a model for complex coding, agentic work (long tasks with tools), expert knowledge, mathematics and front-end builds. In the chat it has automatic thinking and search, web dev, deep research, slides and translation modes, plus tools for image generation and a code interpreter. There are no numbers to back these claims yet.
Where to use Qwen4
- Qwen Studio (chat.qwen.ai): the simplest route. Create an account, pick “Qwen4-Max-Preview” from the list and type in your language. Without an account the model is not available.
- Qwen app: early access on your phone.
- Qoder and QwenWork: Alibaba’s coding environment and work app, also with early access.
- API through Token Plan: from $6 a month ($3 for new users until 30 November), Alibaba Cloud account, Singapore region. It works with OpenAI-compatible tools such as Qwen Code, base URL
https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1.
Two traps from the documentation: Qwen Code’s default output limit for this model is 32,768 tokens, and from Qwen4-Max on, max_tokens counts thinking and the answer together, so a limit that is too low can leave you with no answer at all.
Price
During the preview Qwen4-Max has no per-token price. You buy a Token Plan (Lite $6, Essential $10, Standard $18, Pro $43 a month) and get a credit pool for all models in the plan. Until 25 November Qwen4-Max uses credits at a discount: as low as 20% of the standard rate, and only 8% between 14:00 and 00:00 UTC. How many credits one request takes is visible only in the console.
Qwen4-Max against the competition
| Model | API price ($ per 1M in/out) | AA index | Context |
|---|---|---|---|
| Qwen4-Max-Preview | Token Plan subscription only | not listed | 1M |
| Claude Fable 5.2 | 8 / 40 | 65 (max) | 1M |
| Gemini 4 Carbon | 2 / 10 (until 30 Nov) | 60 (high) | 1M |
| GPT-6 Astra | 10 / 50 | 57 (max) | 1.05M |
| GPT-6.1 Sol | 2 / 10 | 55 (max) | 1.05M |
| Qwen3.8 Max | 2 / 6 | 45 (27 Sep) | 1M |
Prices from OpenRouter, Artificial Analysis index, 11 November 2026.
Until there are numbers, Qwen4-Max cannot honestly be placed higher or lower in this table. It is the cheapest way in (a $3-6 subscription), but the least transparent bill.
What comes next for Qwen4
Alibaba presents Qwen4 as a whole family, but apart from Max it has given no names or dates. In previous generations Max was followed by cheaper Plus and Flash models and by smaller open-weight models you can run on your own computer. When the next Qwen4 models appear, I will add them here.
Launch report: Qwen4-Max preview: what Alibaba showed.
Sources
- Qwen on X, announcement of 11 November 2026.
- Token Plan (Personal Edition) and Model Studio model list.
- Qwen Code in Model Studio.
- Qwen Studio: model list (11 November 2026).
- OpenRouter: competitor prices (11 November 2026).
- Artificial Analysis: intelligence index.
More on this from QUANTSLANT
news·Qwen4-Max Preview: Alibaba opens Qwen4, no benchmarks, two weeks onlyQwen4-Max Preview is live in Qwen Studio, the Qwen app, Qoder and via the Token Plan API. 1M context, preview until 25 Nov, 8% night rate, no benchmarks yet.