How Much Does Qwen 4 Max Cost? Token Plan Tiers and Limits
Qwen 4 Max has no public API price. You get it through Alibaba's token plan. Here are the tiers, credits, preview discounts and the catches to know first.
For Engineering, Finance and Procurement

How much does Qwen 4 Max cost?
Qwen 4 Max has no per-token API price and no downloadable weights. The only way to use it is Alibaba's token plan, a monthly subscription that gives you a pool of credits. Based on a reviewer's walkthrough, the plan has four tiers, with the middle one at $10 a month and the top one at $43 (cut from $68 shortly before launch). The preview was also billed at a steep discount to the normal credit rate. Check Alibaba's current pricing page before you buy, because these numbers have been moving.
TL;DR
- Qwen 4 Max is sold only through a subscription, not as pay-per-token API access and not as open weights.
- The token plan has four tiers with monthly credit pools of 20,000, 45,000, 80,000 and 200,000 credits, according to the reviewer who tested it.
- During the preview, Qwen 4 Max used about 20% of the normal credit rate, and about 8% at night China time, so credits went much further than they will later.
- Alibaba does not publish credits per request, and you can only see usage in its console, so you cannot predict your bill-equivalent in advance.
- The plan is sold in the Singapore region only, the individual plan reportedly cannot be canceled, and it is meant for interactive coding tools, not automation scripts.
- The $10 tier is a low-risk way to try it. Paying for the top tier just for a short preview is hard to justify.
What is Qwen 4 Max and why is the pricing unusual?
Qwen 4 Max is Alibaba's new flagship model for coding and agentic work (tasks where the AI plans and carries out many steps on its own). It has a 1 million token context window, accepts text, images and video as input, and has thinking turned on by default. One creator reported that Alibaba's team describes it as a 2.4 trillion parameter model with about 70 billion active per token. Smaller Qwen 4 models are said to be coming as open source.
The launch was unusual. It arrived as a preview on November 11 with no benchmark table, no open weights and no API price. That means you cannot call it the way you call most models, paying per million tokens. You buy a plan and spend credits.
What are the token plan tiers?
One reviewer listed four tiers, and the credit pools are 20,000, 45,000, 80,000 and 200,000 credits a month. The prices he gave were $6, $10, $18 and $43. (His top price is garbled in the transcript, but he separately says the top tier was cut from $68 to $43, so that is the number I use.) Matching them in order gives this:
| Tier price (per month) | Monthly credits | Parallel agents listed |
|---|---|---|
| $6 | 20,000 | Not stated |
| $10 | 45,000 | 2 to 3 |
| $18 | 80,000 | Not stated |
| $43 | 200,000 | 6 to 8 |
A second creator mentioned a plan at about $3 a month and said he was on a "standard" plan. The two reviewers do not match on entry pricing, so tier names and prices may differ by account or have changed since launch. Treat the table as a snapshot, not a quote.
The plan is also not Qwen-only. It reportedly includes Qwen 3.8, GLM 5.3 and the DeepSeek V4 models. For comparison, Alibaba's older coding plan was $50 a month and does not include Qwen 4 Max.
What did the preview discount actually do?
Two days before launch, Alibaba raised the credits on every tier. During the preview, Qwen 4 Max also drew down credits at only about 20% of the normal rate, and about 8% at night China time.
This explains why the heavy testing went smoothly. One reviewer ran six large app builds at the same time, about 104 million tokens in total (most of them cached reads), and hit no rate limit or quota error. He ran them on the top tier, which lists six to eight simultaneous agents. The other reviewer said he used a cheaper plan daily "without meeting any rate limits or throttling."
Do not assume that experience carries over once the discount ends. If the normal rate is five times higher, a pool that felt endless would feel small.
What are the catches?
Read these before paying:
- No published cost per request. Alibaba does not say how many credits a request uses. You see usage only in its console, after the fact.
- Singapore region only. The plan is sold there, which may matter for data policies at your company.
- No cancellation on the individual plan. According to the reviewer, you cannot cancel it.
- Interactive use only. The terms reportedly say the plan is for interactive use in coding tools, not automation scripts. If you wanted to wire it into a scheduled job, that is a problem.
- The model has an end date. One review says the preview model retires on November 25, so you may be paying for whatever comes next. The second video describes the model as out of preview, so confirm the current status.
- Output limits in some tools. In Open Code, the output limit for this model defaults to 32,000 tokens, and the reviewer raised it before running his tests.
Is Qwen 4 Max worth paying for?
It depends on what you want. The two reviewers disagree, and both are single-person tests, not independent benchmarks.
The more cautious reviewer scored it 30.6 out of 60 across six one-prompt app builds, rank 11 of the 14 models he had tested. It built a strong room designer and keyboard configurator but missed two of three consoles in an emulator and shipped a dead start menu in a Pokémon-style game. He also noted it sometimes claimed things were "verified" that it never checked. His verdict: a decent preview, not a new leader. A much cheaper model, DeepSeek V4.1 Flash at 30 cents per million input tokens, scored slightly higher on his bench.
The other reviewer was far more enthusiastic. He praised a vision-to-simulation test, a bug hunt in a full-stack app, and a research briefing made with web search and other tools. He called it frontier-grade for the price.
A fair reading: it is strong at research-heavy, tool-heavy work and weaker at getting a complete app right in one pass. Keep a human or a second model reviewing the output.
Which tier should you pick?
- Trying it out: the $10 tier. Its 2 to 3 parallel agents is enough for one person in one editor.
- Running several agents at once: only the top tier lists 6 to 8. Be sure you need that.
- Planning around a fixed budget: hold off, since you cannot forecast credits per request.
- Scripts and automation: this plan is not meant for it, per the terms as reported.
Could you track AI plan spend in your own tool?
If you juggle several AI subscriptions with different credit pools, renewal dates and rules, a spreadsheet gets stale quickly. With Remy, you can describe a small internal tracker in plain English: each plan, its tier, monthly credits, who uses it, the renewal date, and a notes field for catches like region or cancellation terms. Remy builds and deploys it as a full-stack app you own. Because plans like this report usage only in their own console, you could add a form where people log their weekly credit readings and see which subscriptions are worth keeping.
Frequently Asked Questions
Does Qwen 4 Max have an API price?
Not at launch. The reviewers say there was no API price and no open weights. Access came only through the token plan subscription, and credits are drawn down per request at a rate Alibaba does not publish.
Is Qwen 4 Max free?
No. Even the cheapest plans cost a monthly fee, and one reviewer mentioned a plan at about $3 a month. The preview discount lowered how many credits each request used, but it did not make the model free.
Can I run Qwen 4 Max locally?
Not today, because there are no open weights. Alibaba's team reportedly plans to release smaller Qwen 4 models as open source, but that would be different models, not Max itself.
Can I cancel the token plan?
According to one reviewer, the individual plan cannot be canceled. Read Alibaba's terms before paying, and treat the $10 tier as the most you would risk on a trial.
How we source, write, and correct our articles. Editorial standards
Related articles

How to Review AI Coding Tool Usage Before Renewal

Is Business AI Spending Flat? What Ramp's Data Shows

Genspark Media Studio: Models, Credits and Free Quota Explained
