LONG AI
Home › Blog › Qwen4 Preview Listed: Subscription-Only,

Qwen4 Preview Listed: Subscription-Only, Retiring Nov 25

Category:资讯 · 6 min read · 2026-11-12

At 10:00 on the morning of Nov 11, Alibaba Cloud Bailian's "model listing and updates" page gained a row: qwen4-max-preview. This is the first time the Qwen4 generation has an official docs page - the Oct 31 practitioner warm-up and the Nov 4 repo/model leak were only rumors before this.

But the same announcement carries two easily-skipped restrictions: Token Plan only, no pay-as-you-go; the preview runs 11-11 10:00 to 11-25 24:00, retiring after. In other words, it's a trial window that closes in fourteen days, not a new model to wire into production.

1. Official source and dates

2. What actually changed

Self-made: left Qwen4-Max-Preview timeline Oct 31 warm-up, Nov 4 leak, Nov 11 10:00 listed, Nov 25 24:00 retired; right new vs old flagship comparison with 1M context, 393K max output, Token Plan only, plaza search for qwen4 empty, and qwen3.8-max at 0.66/1.98 yuan/M
Fig 1 - Self-made: left the preview timeline, right the new preview vs still-on-sale qwen3.8-max. Official price and spec from Bailian docs (2026-11-11), plaza price from our API test (2026-11-12 09:47).

3. Three things for ordinary users

First, can you use it. Yes, but the gate moved. Your pay-as-you-go Bailian key won't work; you must subscribe to Token Plan (personal or team; monthly, quarterly, or annual), then generate a plan-specific key in the console. (Subscription contracts differ from pay-as-you-go; read the cancellation terms first.)

Second, you can't price it per token. The official side gives no per-unit price for this model; Token Plan meters everything in Credits. So you can't compute "how much per million tokens" - only "how many jobs per package." That's completely different from qwen-max's "input cost X, output cost Y" intuition - if you want to compare prices, there's nothing to compare against here.

Third, should you switch. Don't switch it into production. It retires at 24:00 on Nov 25; write the model name into code and it's guaranteed to error two weeks later, and the official side says "no changes during the window" - not even a patch window.

One more biting detail: from qwen4-max-preview on, max_tokens and max_completion_tokens become equivalent, both capping the "thinking + answer" total. If old code only reserves budget by answer length, enabling thinking will truncate early. The official docs call this out separately, meaning users have already tripped on it.

4. Our verification: we don't have it yet

At 09:47 on 2026-11-12, we pulled our plaza API: 184 in-stock models, 280 listings. The Qwen family has 9 on sale - qwen3-coder-next, qwen3.6-plus, qwen3.7-max, qwen3.7-plus, qwen3.8-2.4t-a95b, qwen3.8-27b, qwen3.8-flash, qwen3.8-max, qwen3.8-omni-flash - no qwen4 model at all. Typing "qwen4" into the plaza search returns "no matching models."

Plaza search for qwen4 shows no matching models, top shows 184 models, 280 listings, 1.8 off average
Fig 2 - Our plaza test (2026-11-12 09:47): "qwen4" returns nothing, page scope 184 models, 280 listings.

For price comparison, the same-family current flagship qwen3.8-max is more relevant. It has 1 verified seller today, labeled 95% off, page shows ¥0.6597/M in and ¥1.98/M out, official list ¥12.00 / ¥36.00. The math: 12.00 × (1 − 94.5%) = 0.6597, 36.00 × (1 − 94.5%) = 1.98. That's the buyer's actual price, matching the page.

5. Our advice

  1. To try it, buy the lowest package - verification only, no production. Finish your test scenarios before Nov 25, especially long context (near 1M) and video input, which pay-as-you-go older models can't give you.
  2. Production stays on qwen3.8-max (plaza ¥0.66 / ¥1.98 per M, same-gen flagship, on sale long-term); budget-sensitive use qwen3.8-flash or qwen3.7-plus.
  3. Watch two things when integrating: preserve_thinking is on by default, so multi-turn must pass the full reasoning_content from history assistant messages back, not merge it into content; plus the max_tokens budget merge above.
  4. Run a regression on the afternoon of Nov 25, confirming neither code nor monitoring depends on this preview model - the retirement is a hard cutoff, not a "maybe extended."

Scope note

Official info from the three Alibaba Cloud help pages (model page updated 2026-11-11 20:44), spec, dates, and restrictions verbatim. Our availability from the /api/v1/model-plaza API and plaza page test (2026-11-12 09:47), page scope. CNY prices are plaza actual prices, math shown in the body. This article did not test Qwen4-Max-Preview's speed or quality - no pay-as-you-go access and not our in-stock supply, so no "who's stronger" judgment and no vendor self-reported scores relayed. Images are our screenshots and self-made. For billing questions, see how tokens add up.

相关文章

GPT-6.2 Cuts Prices 20%: What the AI Price War Means

GPT-6.2 cuts list prices 20%, yet platform discounts run deeper: gpt-6-astra trades at under 1% of list. What this week's AI price war means for buyer

What Is Prompt Caching? How Cache Tokens Cut API Bills

Prompt caching stores a repeated prompt prefix and bills cache reads at 5-12% of normal input. Real prices, hit conditions, and how to cut your API bi

All 12 Claude Models Compared: Price and Best Use

Twelve Claude versions are on sale. The newest opus-5-5 costs $8.91 a month; the oldest opus-4-6 costs $49.50. Full table and a pick for each workload

Which AI Model Writes Code Best? Speed and Price Tested

gpt-5.6-terra and deepseek-v4.1-flash win on price and reliability; gemini-3.8-flash is fastest. Effective prices, measured speed, and a three-tier pi

One API Key, Every Model: How to Integrate a Multi-Model Marketplace

Point your base URL at the platform and one key covers all 138 on-sale models across 51 vendors, in both OpenAI and Anthropic request formats. Copy-pa