Playbooks · Coding tasks
Qwen4-Max Preview: 14 days, then it retires
Written November 11, 2026
Alibaba's first callable Qwen 4 tier opened on November 11 with a 1M-token context and a 393,216-token output ceiling — and it retires at midnight on November 25. There are no benchmarks, no model card, no parameter count, and no per-token price. The expiry date is the fact that matters most.
The window is the product
Alibaba's Model Studio documentation states the window plainly: the preview runs from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8); no changes will be made during this period, and the model will be retired after the preview ends. A frozen preview is a specific kind of offer: a behavior found irritating on day two will still be there on day thirteen, and reporting it buys nothing until a successor exists. That is the correct trade for an evaluation build and the wrong trade for a dependency. Anything built on this model needs a fallback for November 26, decided before the first call, not after the last one.
Two doors, same fourteen days
There are two ways to reach the model, and they are not the same access. The API route runs through Token Plan only: a monthly credit subscription, Singapore region, with a dedicated base URL at token-plan.ap-southeast-1.maas.aliyuncs.com, so an ordinary Model Studio key does not open it. Pay-as-you-go is explicitly not supported. Personal Edition tiers start at $6 for 20,000 credits, with top-up bundles at $15 per 20,000 credits, and credits are consumed at a rate Alibaba says is set dynamically by model type, token usage, thinking mode, and tool calls — so the only place to learn the real rate is the usage detail in the console. The plan's own terms restrict it to interactive use in coding and agent tools, not automated batch scripts: a preview that may not be batched is a preview to evaluate by hand.
The second door is Qwen Chat, where the same model turned up on November 11: no subscription, no dedicated base URL, no API key, no per-token meter visible to the user. That moves the model from the procurement conversation into the evaluation conversation — but the expiry is identical on both sides, 24:00 on November 25. The chat door feels permanent because chat products usually are. This one is not.
What the model offers
Per Alibaba's documentation, the model carries a 1,000,000-token context and a maximum output of 393,216 tokens, shared between thinking and the final answer. It accepts text, images, and video, and supports function calling, web search, built-in tools, and structured outputs, with compatibility for the OpenAI Chat Completions and Responses APIs and the Anthropic Messages API. The model ID is qwen4-max-preview.
Three settings decide whether an answer is worth reading. reasoning_effort takes low, medium, or xhigh, and defaults to xhigh — the expensive end of the trade, better hard-prompt output with worse latency on easy ones. Starting with this model, max_tokens is equivalent to max_completion_tokens, and both cover thinking plus the final answer, where every earlier Qwen model applied the cap to the answer alone. A request carried over from an earlier Qwen model with max_tokens set to 4,096 now buys 4,096 tokens of thinking and whatever is left — on a hard prompt, that can be nothing, and the failure is a truncated response rather than an error to catch. preserve_thinking is on by default in Chat Completions, so prior thinking comes back in the assistant messages' reasoning_content field; concatenating that into content and resending it as history is the wrong shape and will degrade the next turn. reasoning_effort and thinking_budget must not be set together — pick one control.
What is missing
- No benchmarks. No vendor scores and no independent scores exist for this model. Any "Qwen4 beats X" figure circulating now is either a projection or a different model's number wearing this one's name.
- No model card, no parameter count, no per-token price. Alibaba publishes no per-token rate; the model is metered in credits against a subscription. A reseller's per-token price is the reseller's price, not Alibaba's, and is not carried here.
- No open weights. This is a closed hosted endpoint for fourteen days, not a downloadable checkpoint, so the licensing questions that apply to a downloadable model do not arise.
- Not Qwen4 Max. The preview precedes a generally available flagship that has no model card, no published per-token price, and no pay-as-you-go endpoint. Alibaba named four tiers at Apsara on September 22 — Max, Flash, Plus, and 27B — and this preview is the only one with an endpoint. Treating the preview's numbers as the flagship's numbers is how a roadmap slide becomes a procurement mistake.
What to do with fourteen days
Fourteen days is enough to learn something specific if the test is decided before starting. Three prompts, run twice each, beats a week of open-ended chatting. First, hand it a long document or a screen recording with a known answer, and check whether it finds a detail planted near the end — a million tokens is a claim about capacity, not about attention, and only a planted needle tells the difference. Then test the reasoning default against the cheaper setting on the same prompt: if xhigh and medium land in the same place, that is latency not worth paying for. Then run the video input on something short and specific, like a UI recording with a known bug in it — video understanding is the least benchmarked capability on any model card. Write down what was learned, with dates and settings. In fifteen days that document is the only asset the preview produces.
For work that has to keep running after November 25, the shelf already holds the fallback: Qwen3.8-Max has been generally available since August 3, 2026, with a 1,000,000-token context and text, image, and video input, at $2.00 per million input tokens and $6.00 per million output. It is the model to build on; the preview is the model to learn from.
Related playbooks
- Antigravity CLI 1.5.5: swarms in preview
- BMad v7 is a preview, not the stable release: what changes, and whether to try it before it ships
- Decide if Copilot Code Review Max is worth it
- Gemini CLI 0.68 preview: extension checks
Sources
- Qwen4-Max-Preview — Alibaba Cloud Model Studio
- Explore Qwen3, Wan2.6 & Latest Models on Model Studio — Alibaba Cloud
- Qwen4-Max Preview in Qwen Chat: a second door, same 14 days — OrcaRouter
- Qwen4-Max Preview vs GPT-6.1 Sol: countdown vs rate card — OrcaRouter
Back to: Coding playbooks hub · Also: Kimi K3.1 Pro weights are out: what 1.56 TB buys · Gemini 4 Carbon is here: $2 in, $10 out