Releases

Qwen4-Max

MiniRouter and Qwen4-Max by Alibaba, separated by a silver fold of light on black.

Alibaba's next Max flagship brings adjustable reasoning to coding, document analysis and long agent workflows. Start with text, give it the relevant sources, and keep verification in the loop.

Qwen4-Max is available on MiniRouter as . The model page has current pricing, provider routes and API setup.

View Qwen4-Max

A new Max model for work with several steps

Qwen4-Max follows the Qwen3.8-Max generation. Alibaba positions it around coding, office work and agents that need to plan, use tools and inspect what they produce. Its million-token context gives you room to bring related source material into one request.

That makes a code review with requirements and test output a useful first experiment. For document work, supply extracted text with stable source labels and ask the model to separate supported conclusions from open questions. MiniRouter currently accepts text input for this model.

Alibaba model guide · November 15 gateway announcement

Keep a check between reasoning and action

A model can propose the next step; your application still owns execution. This pattern works for a repository assistant, a document review or a reporting workflow.

  1. 01

    Supply context

    Label the code, requirements and source documents. Define what a correct result must include.

    A question with evidence

  2. 02

    Reason and use tools

    Let the model propose a plan or request a tool. Your application decides which calls may run.

    A proposed answer or action

  3. 03

    Check the result

    Run tests, check cited passages and validate tool results. Return failures for another attempt.

    An accepted result or a revision

Suggested application workflow by MiniRouter, informed by Alibaba's coding and tool-use guidance. This is a design pattern, not a measured benchmark.

Use tools with narrow permissions. A review assistant may need to read files and run tests; it does not need permission to merge a change. Preserve failed checks in the conversation so the next attempt can address the specific failure.

Choose effort with an acceptance test

Qwen4-Max offers low, high and xhigh effort. Start with a small set of real tasks whose answers you can judge, then compare settings on the same inputs. Record whether the result passed, how long it took and its billed usage.

TaskAcceptance
Code reviewA reproducible failure case and a test for the proposed fix
SourcesEach conclusion points to an actual source passage
ToolsArguments are valid and the tool result confirms success

Keep effort and a separate thinking budget as alternative controls. A short requested answer does not imply a small reasoning bill: MiniRouter can reserve the full output allowance for reasoning requests. Keep enough balance for the reservation and inspect usage after completion.

What to revisit when switching

Run Qwen4-Max beside your current Qwen3.8-Max workflow before replacing it. Check answer format, tool arguments, source fidelity and latency. A newer model can behave differently even when the request shape stays the same.

The catalog lists a 384,000-token maximum output, while Alibaba's own guide describes different limits by thinking mode. Treat the limit as route metadata, not a target length. Start well below it. Schema-enforced JSON, image and video input are not exposed on this MiniRouter route.

See current MiniRouter prices and limits →

Start with a text review

Set MINIROUTER_KEY to an API key. This request asks for failure cases you can inspect before connecting any tools.

Qwen4-Max · Chat Completions
curl https://api.minirouter.sh/v1/chat/completions \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H 'Content-Type: application/json' \
  --data '{
    "model": "alibaba/qwen4-max",
    "reasoning_effort": "low",
    "max_completion_tokens": 4096,
    "messages": [{
      "role": "user",
      "content": "Review this retry policy: retry every failed payment three times. Identify failure cases and propose acceptance tests."
    }]
  }'

Model setup and reasoning controls →

Keep exploring

Compare published results in the model rankings and coding benchmarks, or read more model releases.

For MiniRouter product updates, visit the changelog.

Sources

  1. Qwen4-Max model guide and pricing

    Published Nov 16, 2026

    Read at Alibaba Cloud ↗
  2. Qwen4-Max on AI Gateway

    Published Nov 15, 2026

    Read at Vercel ↗
  3. Qwen4 Max evaluation

    Published Nov 16, 2026

    Read at Artificial Analysis ↗

Editorial reference revised 2026-11-16. Catalog status comes from the generated routing snapshot.

Every model, one key.