Qwen4-Max
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.
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.
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.
- 01
Supply context
Label the code, requirements and source documents. Define what a correct result must include.
A question with evidence
- 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
- 03
Check the result
Run tests, check cited passages and validate tool results. Return failures for another attempt.
An accepted result or a revision
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.
| Task | Acceptance |
|---|---|
| Code review | A reproducible failure case and a test for the proposed fix |
| Sources | Each conclusion points to an actual source passage |
| Tools | Arguments 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.
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.
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."
}]
}'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
- Read at Alibaba Cloud ↗
Qwen4-Max model guide and pricing
Published Nov 16, 2026
- Read at Vercel ↗
Qwen4-Max on AI Gateway
Published Nov 15, 2026
- Read at Artificial Analysis ↗
Qwen4 Max evaluation
Published Nov 16, 2026
Editorial reference revised 2026-11-16. Catalog status comes from the generated routing snapshot.