
Qwen4 Max Preview: Official Specs, Access, and Limits
By Nathan Cole
MyClaw Editorial
MyClaw
Run Best-in-Class AI Agents Now
Run OpenClaw or Hermes Agent on managed MyClaw hosting, so your AI agent stays online, updated, and ready for real work.
AI Takeaway
- What is Qwen4 Max Preview? The official name is Qwen4-Max-Preview (
qwen4-max-preview; aliasqwen4.0-max-preview), a next-generation flagship multimodal preview, not Qwen4-Max, Qwen3.8-Max, Qwen3-Max, or QwQ. - How long does the preview last? Official access is 10:00 on 11 November 2026 to 24:00 on 25 November 2026 (UTC+8). The model is then retired, with no changes during the window and no official final-release date as of 13 November 2026.
- What official specs and limits matter in practice? Official docs list a 1M context and a 393,216-token output cap shared by thinking and the answer, with text, image, and video in and text out. Thinking defaults to
xhigh. - How do you access it, and what does official pricing look like? Official access is Token Plan only, not pay-as-you-go. Billing is in Credits, with no published per-token price.
- Where does MyClaw fit? Use MyClaw to plan the prompt, context, and workflow around this preview. Confirm this exact model on official Token Plan or Qwen surfaces.
What Qwen4 Max Preview Is — and What It Is Not
Qwen4 Max Preview is Qwen’s next flagship multimodal model in a time-boxed official preview, not a generally available Qwen4-Max release. The official name is Qwen4-Max-Preview. The formal model ID is qwen4-max-preview, with the alias qwen4.0-max-preview; official Token Plan bills show the alias. The vendor surfaces are Qwen, Alibaba Cloud Model Studio, and QwenCloud.
Official documentation sets the preview from 10:00 on 11 November 2026 to 24:00 on 25 November 2026 (UTC+8). In that window the model will not change, and official docs say it will be retired when the preview ends. As of 13 November 2026, they do not publish a date for a final Qwen4-Max.
Official product copy describes text generation, image and video understanding, and deep reasoning, and names coding, office productivity, frontend design, and complex agentic tasks as the intended jobs. It also supports web search, tool calling, structured outputs, and APIs compatible with OpenAI Chat Completions, Responses, and Anthropic Messages.
Do not collapse this preview with earlier Max-class models. It is not Qwen3.8-Max, not the older Qwen3-Max or Qwen3.5-Max lines, and not QwQ or QVQ. It is also not Qwen3.8-Flash-Next, an official open-weight architecture preview of Qwen4 and a different product, compared in Qwen 3.8 Flash-Next vs Qwen 3.8 27B. Official materials for this preview do not publish a parameter count or downloadable weights.

Official Specs at a Glance
Use the official specifications below as of 13 November 2026. This is a capability snapshot, not a quality ranking.
| Item | Official value (as of 13 November 2026) |
|---|---|
| Official name / IDs | Qwen4-Max-Preview; qwen4-max-preview; alias qwen4.0-max-preview |
| Status | Preview only, 11–25 November 2026 (UTC+8), then retired; no official GA date |
| Context | 1M tokens |
| Max output | 393,216 tokens, shared by thinking and the final answer |
| Input / output | Text, image, and video in; text out |
| Visual limits | 16M pixels per image; video up to 2 hours and 2 GB; up to 2,048 URL images and 64 videos. Base64 image cap: not officially specified |
| Thinking | Hybrid mode, enabled by default; low / medium / xhigh (default xhigh). Thinking budget: not officially specified |
| Tools and output format | Function calling, web search, built-in tools; JSON Object and JSON Schema |
| API compatibility | OpenAI Chat Completions, Responses, and Anthropic Messages, via a Token Plan key and dedicated endpoint |
| Official pricing unit | Token Plan Credits only. No published pay-as-you-go input, output, or cache price |
| Size / weights | Not officially published |
Starting with this model, official OpenAI Chat Completions notes treat max_tokens as equivalent to max_completion_tokens. Both count thinking and the final answer together, so a limit sized for the answer alone may truncate output. Do not set reasoning_effort and thinking_budget together. Official QwenCloud tables mark the thinking budget and the Base64 image cap as not specified upstream; international documentation lists a single max-images figure of 2,048 and does not split a Base64 cap. Leave those cells unspecified.
As of 13 November 2026, official benchmark results for this preview remain unpublished in the official Model Studio, QwenCloud, and Qwen sources reviewed for this article.
Practical Uses and Official Limits
Official copy points this preview at coding and agent workflows, office and knowledge work, frontend or UI-oriented tasks, and screenshot, document, or video understanding in the same text-out model. That describes the intended jobs, not a measured result on your brief.
Access is Token Plan only, and official docs retire the model at the end of the preview window. With no published replacement date, don't treat qwen4-max-preview as a durable production dependency. Finish time-sensitive work inside the official window, or keep a fallback model ready for anything that has to continue after it.
Size the output budget for thinking plus the answer; a tight answer-only limit can get cut short, because official docs share that cap between reasoning and the reply. Output is text only — this is not documented as an Omni speech model. The Base64 image cap is not officially specified. Official Token Plan terms limit use to interactive work in coding and agent tools, not automation scripts, custom application backends, or non-interactive batch jobs. This preview is not a self-host option. For a long coding or document job, plan around the official 1M context rather than a short chat window.
If the work cannot depend on a retiring preview, compare the previous official Max-class option in Qwen 3.8 Max vs Kimi K3. Confirm current access on official surfaces; that post is not a stand-in for this preview.
How Official Access and Pricing Work
Official access is through Token Plan only — Personal (called Individual on QwenCloud) or Team — with a dedicated Token Plan API key. Pay-as-you-go is not supported, and there is no published per-token input, output, or cache price for qwen4-max-preview. Set the model ID to qwen4-max-preview, or the documented alias qwen4.0-max-preview. International Token Plan documentation places the service in the Singapore region, with Global deployment and terms that prompts and outputs involve cross-border data transfer. China-site documentation lists a different Base URL; do not treat the international dollar prices below as China-site pricing.
As of 13 November 2026, official international and QwenCloud Personal / Individual list prices are: Lite $6 per month (limited-time new-user $3) for 20,000 Credits; Essential $10 (new-user $5) for 45,000 Credits; Standard $18 for 80,000 Credits; and Pro $43 for 200,000 Credits. An extra pack is $15 per month for 20,000 Credits, requires an active subscription, and allows up to five packs. Official docs say the new-user price runs through 23:59:59 on 30 November 2026 (UTC+8) and applies only to accounts that have never purchased a Token Plan Personal subscription. Unused monthly quota does not carry over.
Official Token Plan notes also describe a limited-time Credit promotion the vendor may change: consumption as low as 20% of the standard Credit rate (5x usage), and calls between 22:00 and 08:00 (UTC+8) at 8% of the standard rate. That is a Credit-consumption rate, not a published per-token dollar price. Official docs say Credits per request vary with model type, token use, thinking mode, and tool calls; check actual burn in console usage details.
Plan the Work in MyClaw

MyClaw is a hosted AI team that works with your tools around the clock. Use it to plan the prompt, context pack, and agent workflow for this preview window, then refine that plan as work rather than as a one-off chat demo.
- Open the MyClaw homepage and choose Get Started.
- Write the actual brief — a coding task, a long-document pass, or a multimodal review — and size the context and output budget to the official limits in the table above, including room for thinking as well as the final answer.
- Use MyClaw to refine and run the surrounding workflow: iterate on the prompt, the attached context, and the agent steps until the brief is ready to use.
If the job includes screenshots, documents, or video, carry those official visual limits into the brief so the plan matches how the preview is documented. If you need this exact model during the official preview window, use official Token Plan or Qwen surfaces and confirm the current official model list.
Conclusion
Qwen4 Max Preview is a Token Plan-only flagship preview with documented official specs and a hard end date of 25 November 2026 (UTC+8). Use official access if you need this exact model before then, and don't plan as if a final Qwen4-Max ID already exists. Use MyClaw to plan and run the surrounding workflow. Confirm live official docs if the window or Credit terms change, and keep long-running work on a path you can still use after the preview is retired.
Skip the Setup, Run Best-in-Class AI Agents Now
Launch a managed OpenClaw or Hermes Agent workspace in minutes, with always-on hosting, updates, and support handled by MyClaw.