Alibaba
Alibaba

Qwen4 Max

Release Date 2026-11-161M IN · 131K OUTProprietary ModelAPI available
Overall3.0Top 27%
AutomationResearching
Docs & officeResearching
CodingResearching

Qwen4 Max is Alibaba's flagship AI model in the Qwen family. It understands text, images and video together, and it thinks longer before answering when a task calls for it. The company says it improves on its predecessor, Qwen3.8 Max, in an in-house evaluation of demanding professional tasks. According to Alibaba, it also uses noticeably fewer tokens and cuts the average cost per task by a large margin. The company adds that it activates fewer parameters per token than the previous version at the same overall scale, which it credits for the lower running cost. It is aimed at coding, office productivity, frontend design and multi-step tasks that it carries out on its own, and it supports web search and tool calling. It also reads long documents in one go, and developers can plug it into their own products.

API cost

$1.43IN$4.28OUT

Real-world usage

No.46New entry0.2% share · 156B tokens/wk
156B11/16
SourceOpenRouterLast updated

Output Speed

Very slowper 2,000 tokens
41tok/s
First Output 2.1stotal: 50.9s
SourceOpenRouterLast updated

Expert Performance Evaluation

How strong is it across many tests combined?AA Intelligence Index v5Top 31%#15
42.9

More about AA Intelligence Index v5

40.4%41.3%21.3%66.7%

More about Tax Agent Bench

Can it build a working web app from a plain-language spec?Vibe Code Bench v1.1Top 17%#15
86.7%

More about Vibe Code Bench v1.1

See each evaluation’s detail page for its source and last updateData sources & removal requests