Qwen4-Max Preview: Where to Use It and What Isn’t Announced

Last checked 2026-11-11 17:28 Beijing time (UTC+8), Token Plan docs; the other sources were read at about 13:00. Corrected to match the official page as read on 11-11 at 17:28 (night-discount wording).

Short version: Qwen4-Max is in preview, not released. You can try it on Alibaba’s Token Plan and in Qoder today. Alibaba has not published an API price, a context window, benchmarks, a parameter count or open weights. The Token Plan preview ends on 11-25 at 24:00 Beijing time, and the model is taken offline afterwards. Treat it as a two-week trial.

What was announced, and when

• 11-11 10:01 Beijing time (02:01 UTC): the official Qwen account (@Alibaba_Qwen) posted “Qwen4 is coming.” and “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.” (post)
• 11-11 10:31 Beijing time (02:31 UTC): Qoder’s official account said its anonymous “Qoduck” model, which had been a Qwen test build for four days, “upgrades to Qwen4-Max-Preview” and is “live first on Qoder: every product, every plan, Teams and Enterprise included.” (post)

So Qoduck was a Qwen test build, and Qoder now calls it Qwen4-Max-Preview. Qwen describes the goal as “deeper knowledge”, “richer world understanding” and “stronger coding and agentic capabilities”. That is Qwen’s wording, not a measured result.

Where you can use it now

Own chart, built from the two official posts and the Token Plan docs.

• Token Plan is the only place with a documented model ID, dates and a Credits discount. The China-platform docs list qwen4-max-preview (alias qwen4.0-max-preview) with these capabilities: reasoning, vision understanding and text generation.
• Qoder is the only other place with a stated offer, quoted below.
• QwenWork, Qwen APP and QwenStudio are named in Qwen’s post as “early access”. I found no official page with Qwen4 quota or price for any of them.

Token Plan: tiers and the preview discount

Figures from the Token Plan Personal docs, in RMB.

The numbers above come from the Token Plan Personal docs on the China platform (the Qwen AI platform, platform.qianwenai.com), in RMB. What the docs say about Qwen4-Max:

• Preview window: 11-11 10:00 to 11-25 24:00 Beijing time. “No changes during that period; taken offline after the preview ends” (our translation).
• During the promotion, Credits use is “as low as 20%” of the standard rate (“equivalent to 5x the usage”, in the docs’ words).
• From 22:00 to 08:00 Beijing time each night there is a further 40% rate, which the docs describe as 8% of the original standard (our translation of the Chinese page; see the official page).
• New-user prices exist only for Lite (¥19.5 instead of ¥39) and Essential (¥39.5 instead of ¥79). The page lists none for Standard or Pro.
• The docs do not say how many Credits one Qwen4-Max request costs, so none of these figures tells you how many tasks a plan covers.

Two caveats the docs also state: the personal edition currently does not support cancelling a subscription, and model inputs and outputs are used for service improvement and model optimisation while you use the plan. Use is limited to interactive use inside coding and agent tools, not scripts or backends. Read those lines before you buy a plan only for a two-week preview.

Two Token Plan sites. Qwen’s post links to qwencloud.com/pricing/token-plan, the international site. Its docs say Token Plan can only be bought and used in the Singapore region, with its own endpoint and keys. When I read that page, its static text had no Qwen4 price line and the prices load dynamically, so I could not read them. I’m not mixing its USD prices with the RMB table above.

Token Plan works with Claude Code through the Anthropic protocol, according to its docs. If you want to see how other Chinese models are wired into Claude Code, see DeepSeek API for Claude Code, with Kimi and GLM as alternatives.

Qoder: “80% off, 0.1x credits”

Qoder’s post says, word for word: “Limited time: 80% off, 0.1x credits.”

Those two phrases do not describe the same price:

• 80% off means you pay 20% of the normal price, which is 0.2x.
• 0.1x credits means you pay one tenth, which is 90% off.

Qoder’s later post at 11:03 Beijing time (03:03 UTC) says only “Try it now at 80% off for a limited time!”, with no 0.1x. Qoder does not explain which figure applies, and does not give an end date in either post. Qoder’s pricing page, which I read at about the same time, lists plans and credit packs but no per-model rates, so it can’t settle this. We are not merging the two, and we are not adding Token Plan’s discount to Qoder’s: they are different products with different credits. If you use Qoder, update to the latest version, pick Qwen4-Max-Preview, and compare your credit balance before and after a small task before you start a long one. That is our suggestion, not something Qoder says.

What has not been announced

“Not announced” means nothing on the official pages we checked at about 13:00 Beijing time on 11-11. It can change within hours.

• Pay-as-you-go API price
• Context window
• Benchmark results (so we cannot say how it compares with other models)
• Open weights
• Parameter count
• Date and name of the final release
• Credits per request on Token Plan
• End date of Qoder’s offer
• Quota or price on QwenWork, Qwen APP and QwenStudio

Alibaba Cloud’s Model Studio pay-as-you-go price page, which I searched for “qwen4” at about 12:55 Beijing time, lists no Qwen4 model.

What we left out on purpose

• “10% of the original price” in Chinese media. Phoenix (ifeng.com) reported on 11-11 11:22 that Token Plan users get a 20% rate, new users another 50%, “at lowest 10% of the original price”. The official docs list the 20% Credits rate and the new-user prices separately, and never print the combined 10%. We don’t repeat it as an official number. Standard and Pro list no new-user price.
• Anything that appears only in secondhand posts and not on an official page. We only report what Alibaba, Qwen or Qoder wrote themselves.

Before you spend anything

1. Remember the end date: 11-25 24:00 Beijing time. Don’t build a workflow around a model that goes offline.
2. If you want the night discount, the window is 22:00 to 08:00 Beijing time, as the docs state.
3. Don’t compare Qwen4-Max with other models yet. There are no published benchmarks.
4. Check the official pages again before paying. Token Plan’s docs say the promotion can be changed, and Qoder’s offer is “limited time” with no end date.

For how Chinese-model API prices compare today, see our LLM API pricing comparison, and for another China-model price breakdown, Kimi K3.1 Pro API pricing. We will update this page when Alibaba publishes an API price.

Sources (read 2026-11-11, about 12:45 to 13:00 Beijing time; Token Plan docs re-read at 17:28)

• Qwen: @Alibaba_Qwen post
• Qoder: @qoder_ai_ide post and follow-up post, 11:03 Beijing time
• Token Plan Personal docs (China platform): platform.qianwenai.com
• Token Plan price-change notice of 11-09: platform.qianwenai.com
• Token Plan docs (international): docs.qwencloud.com
• Qoder pricing page: qoder.com/pricing
• Model Studio pay-as-you-go prices: help.aliyun.com
• Media, attributed only: Phoenix / ifeng.com, 11-11 11:22 Beijing