In short: Qwen4-Max-Preview, the new flagship in Alibaba Cloud's Qwen family, can only be used through a Token Plan subscription until the preview ends at 24:00 on November 25, 2026 (UTC+8); there is no pay-as-you-go price at all. The cheapest way in is a Personal Edition plan from $3 a month at the new-user price, and preview calls draw Credits at as little as 20% of the standard rate, or 8% at night. The catch most people will miss: Token Plan is licensed for interactive use inside coding and agent tools only, so it is not a way to run the model behind your own app.
What you are actually buying
Qwen4-Max-Preview (model ID qwen4-max-preview, alias qwen4.0-max-preview, which is the name that shows up on bills) is a multimodal reasoning model. According to Alibaba Cloud Model Studio's model page, it takes text, images and video as input, supports function calling, web search and built-in tools, returns structured JSON (JSON Object and JSON Schema), and speaks three API dialects: OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. The context window is 1M tokens and the maximum output is 393,216 tokens, a budget that thinking and the final answer share.
The model page is unusually blunt about access: it is "available through Token Plan only. Pay-as-you-go access is not supported." The preview runs from 10:00 on November 11 to 24:00 on November 25, 2026 (UTC+8), Alibaba says nothing will change during that window, and the model is retired when the preview ends. So any money you spend here buys roughly two weeks of access to this specific model, plus whatever else the plan covers for the rest of your subscription month.
The preview window in your time zone
- Ends: November 25, 2026 at 24:00 UTC+8, which is 16:00 UTC, 08:00 in US Pacific (PST), 11:00 in US Eastern (EST), and 01:00 on November 26 in Seoul and Tokyo.
- Night discount window (22:00 to 08:00 UTC+8): 14:00 to 00:00 UTC, 06:00 to 16:00 Pacific, 09:00 to 19:00 Eastern, 23:00 to 09:00 in Seoul and Tokyo. In other words, the "night" discount lands in the US working day.
Token Plan prices, read carefully
Token Plan is Alibaba Cloud Model Studio's subscription for AI coding and agent tools, billed in Credits instead of dollars per token. There are two editions, and both list Qwen4-Max-Preview among their supported models.
Personal Edition (Singapore region only)
- Lite: $6/month for 20,000 Credits, 1-2 concurrent agents. Limited-time new-user price $3/month.
- Essential: $10/month for 45,000 Credits, 2-3 concurrent agents. New-user price $5/month.
- Standard: $18/month for 80,000 Credits, 3-4 concurrent agents.
- Pro: $43/month for 200,000 Credits, 6-8 concurrent agents.
- Extra Bundle: $15 per bundle per month for 20,000 Credits, up to 5 at a time, only on top of an active plan.
The new-user price runs until 23:59:59 on November 30, 2026 (UTC+8) and only applies to accounts that have never bought a Personal Edition subscription. The quotas were raised on November 9, 2026, and the Pro list price was cut at the same time, so older third-party write-ups may show smaller numbers.
Team Edition (per seat)
- Standard seat: $20/seat/month, 25,000 Credits.
- Advanced seat: $75/seat/month, 100,000 Credits.
- Premium seat: $180/seat/month, 250,000 Credits.
- Shared usage pack: $700 for 625,000 Credits, valid one month, drawn on when a seat runs out.
What one Credit really costs
Because every plan is "dollars for Credits", the honest way to compare them is price per 1,000 Credits. I divided each list price by its monthly quota (my own table is among the screenshots at the end of this post):
- Personal Essential at the new-user price: $0.111 per 1,000 Credits, the cheapest Credits Alibaba sells right now.
- Personal Lite at the new-user price: $0.150.
- Personal Pro $0.215, Essential (list) $0.222, Standard $0.225, Lite (list) $0.300.
- Personal Extra Bundle: $0.750, about 2.5 times the Lite list rate and over 3 times Pro.
- Team seats: $0.800 (Standard), $0.750 (Advanced), $0.720 (Premium); the shared pack works out to $1.12.
Three things jump out. First, the new-user Essential plan beats every other option per Credit and its promo outlasts the preview by five days. Second, Extra Bundles are the most expensive way to top up a personal plan; if you keep running out, upgrading a tier is cheaper per Credit (the docs prorate upgrade cost and extra Credits by remaining days in the cycle). Third, Team Edition Credits cost roughly three to seven times Personal Edition Credits. What you get for that is seat management, per-member usage analytics, and a statement that Team conversation data is not used for model training.
What I could not price: Credits per token
Alibaba does not publish a fixed Credit rate per token for Qwen4-Max-Preview. The Personal Edition page says Credits per request depend on model type, token usage, thinking mode and tool calls, and points you to the usage details in the console. The Team page gives one worked example for a different model (qwen3.6-plus: 8,349 input tokens, 40,794 cached tokens and 573 output tokens came to about 3.18 Credits), which shows how cached tokens are counted but cannot be reused as a Qwen4 price. So nobody can honestly tell you "a 1,000-line refactor costs X dollars" on this model yet. The practical move is to run one typical task, read its Credit cost in the console, and multiply.
The preview discount and the night discount
For the preview, Credits consumption is "as low as 20% of the standard rate (5x usage)", and calls between 22:00 and 08:00 (UTC+8) get "an additional 60% off, that is, 8% of the standard rate". Note the wording "as low as": treat 5x as a ceiling, not a guarantee. If your working hours overlap the night window (US daytime, or late evening in East Asia), heavy jobs such as a long repository review are noticeably cheaper there, about 2.5 times more work per Credit than daytime preview calls.
Plan comparison at a glance
| Option | Price | Credits / month | Concurrency or scope | Best for |
|---|---|---|---|---|
| Personal Lite | $6/mo ($3 new-user) | 20,000 | 1-2 concurrent agents | Trying Qwen4-Max-Preview for a few sessions |
| Personal Essential | $10/mo ($5 new-user) | 45,000 | 2-3 concurrent agents | Most solo developers during the preview; cheapest per Credit |
| Personal Pro | $43/mo | 200,000 | 6-8 concurrent agents | Daily heavy agent use with parallel sessions |
| Team Standard seat | $20/seat/mo | 25,000 per seat | One member per seat, admin console | Teams that need seat control and no-training data terms |
All prices from Alibaba Cloud Model Studio's Token Plan docs, checked November 13, 2026.
How to point a coding tool at it
The official quick start is short: subscribe to Token Plan Personal or Team, use the plan's own API key, set the dedicated base URL https://token-plan.ap-southeast-1.maas.aliyuncs.com (each protocol has its own path under it), and set the model ID to qwen4-max-preview. Supported tools listed in the docs include Claude Code, Codex, Qwen Code, Qoder, OpenCode, OpenClaw, Trae and a generic integration guide.
Claude Code example
Alibaba's Claude Code guide uses the Anthropic-compatible path and maps Qwen4-Max-Preview to the "Opus" slot. In ~/.claude/settings.json, the relevant env values are:
"ANTHROPIC_BASE_URL": "https://token-plan.ap-southeast-1.maas.aliyuncs.com/apps/anthropic",
"ANTHROPIC_MODEL": "auto",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "qwen4-max-preview",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.8-flash",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen3.6-flash",
"CLAUDE_CODE_MAX_CONTEXT_TOKENS": "983616"
Then run claude -p "Hello" to check the connection. Note that the guide's default main model is auto, a platform router that picks an underlying model per request. If you specifically want Qwen4-Max-Preview for a session, select the Opus-mapped model or set ANTHROPIC_MODEL to qwen4-max-preview yourself; otherwise you may be paying preview attention to a different model. The guide also says to set CLAUDE_CODE_MAX_OUTPUT_TOKENS=64000 if long output gets truncated.
The max_tokens change that truncates answers
Starting with this model, Alibaba aligned max_tokens with OpenAI's current behavior: on the OpenAI-compatible Chat Completions API, max_tokens now equals max_completion_tokens and caps thinking plus answer together. Thinking is on by default at the xhigh effort level (you can choose low, medium or xhigh via reasoning_effort; do not set it together with thinking_budget). A limit you sized for the answer alone on older Qwen models can now be eaten by reasoning and end in a cut-off reply. Older Qwen models keep their previous behavior, which is why this bites people who just swap the model ID.
Editor's note
Who it's for: developers who already work inside Claude Code, Qwen Code, Codex-style CLIs or similar agents and want to test a 1M-context, multimodal Qwen flagship for two weeks at a low fixed price.
Who it's not for: anyone hoping to call Qwen4-Max-Preview from their own backend, a cron job or a batch pipeline. The Personal Edition terms say the plan "must not be used for automation scripts, custom application backends, or any non-interactive batch call scenarios", and misuse can lead to suspension. There is no pay-as-you-go route for this model, so for production you need a different model.
Gotchas I found in the docs: Personal Edition cannot be unsubscribed once bought, so turn off auto-renewal if you only want the preview month. Unused Credits do not roll over. When both plan quota and bundles run out, the service simply pauses until the next subscription month. Token Plan runs only in the Singapore region in a Global deployment mode, so prompts and outputs cross borders; check that against your company's rules. Multi-turn chats keep preserve_thinking on by default, and the docs ask you to return earlier reasoning in the reasoning_content field, not inside content.
Common errors and fixes: truncated or empty answers usually mean the shared thinking-plus-answer budget is too small; raise max_tokens or lower reasoning_effort. A 401 invalid_api_key in Claude Code has its own entry in Alibaba's Token Plan FAQ; first confirm the tool is pointed at the Token Plan base URL above, not a pay-as-you-go endpoint. If you see a different model in your bills, check whether your tool is still set to auto.
Is it worth it?
For a solo developer, yes, with limits: at the $5 new-user Essential price you get the cheapest Credits Alibaba sells, enough room to put the preview model through real repository work, and the other Qwen, DeepSeek and Zhipu models on the plan for the rest of the month. Do the expensive runs inside the night window, size max_tokens for thinking, and check the console after your first task to learn your real Credit burn. If you need an API you can ship behind a product, skip it: this preview is for hands-on tool use only and disappears on November 25.
If you are weighing other paid model routes, my breakdowns of Gemini 4 Carbon pricing before its intro rates end, OpenRouter's free models, 5.5% fee and BYOK, and running Inkling 2 Small locally or via API cover the pay-per-token side.
FAQ
Can I use Qwen4-Max-Preview with pay-as-you-go billing?
No. The official model page says it is available through Token Plan only and pay-as-you-go access is not supported during the preview.
What happens after November 25, 2026?
Alibaba says the model will be retired when the preview ends at 24:00 UTC+8 on November 25. Your Token Plan subscription continues with the other supported models until it expires; the docs do not say what, if anything, replaces the preview model.
Which plan should I pick just to try it?
If you have never bought Personal Edition, Essential at the $5 new-user price gives the most Credits per dollar ($0.111 per 1,000 in my calculation). Lite at $3 is the smallest bet. Remember Personal Edition cannot be unsubscribed, so manage auto-renewal yourself.
Short version: cheap and generous for two weeks of interactive coding, useless as a production API.
Last verified: 2026-11-13
Sources: Alibaba Cloud Model Studio: qwen4-max-preview; Token Plan overview; Token Plan Personal Edition; Token Plan Team Edition; Claude Code integration guide; API compatibility upgrade notice.
lightmind24 is an independent review site, not affiliated with Alibaba Cloud or Qwen.
Image credits: screenshots of Alibaba Cloud Model Studio documentation pages, plus one cost table made by lightmind24.
댓글
댓글 쓰기