
On this page
Short answer
Alibaba has released Qwen4-Max, the new flagship model in its Qwen family. It reads text, images and video, can work with up to 1 million tokens of context, and is built for coding, office work and long-running AI agents. It is available now through Alibaba's QwenCloud platform, and Alibaba says smaller open-weight Qwen4 models will follow soon.
What happened
Alibaba announced Qwen4-Max on November 16, 2026. According to the company, it is the most capable Qwen model so far: a Mixture-of-Experts model with 2.4 trillion total parameters that activates only 71 billion of them for each token. Alibaba says that is 25% fewer active parameters than Qwen3.8-Max, which helps keep running costs down.
On Alibaba's own Work360 benchmark, which covers tasks from 721 specialized fields, the company says Qwen4-Max beats Qwen3.8-Max overall while using 35% fewer tokens in its high-effort reasoning mode, with the average cost per task dropping by around 60%. Alibaba also reports a 6th place on Code Arena, a real-world coding leaderboard.
Outside tests give a more balanced picture. GIGAZINE reports that Qwen4-Max scored 43 on the Artificial Analysis Intelligence Index: below top US flagships such as Claude Fable 5.2 and Gemini 4 Carbon, but above lighter models such as Claude Haiku 5.5.
QwenCloud's model page lists up to 393K output tokens, built-in tools such as web search and a code interpreter, and compatibility with the OpenAI and Anthropic APIs. Alibaba showed the model building 3D scenes in Blender, helping develop games in Unity, Unreal Engine and Godot, and running a 48-hour agent workflow that, according to the company, found an experimentally validated chip coolant molecule.
Why people are talking about it
Qwen4-Max was one of the top stories on Techmeme, with coverage from outlets such as GIGAZINE and RuntimeWire. Leaderboard projects and developers quickly posted their first rankings, and many people are already asking about the smaller open-weight versions.
What it means if you use AI
- Long chats and big files. A 1 million token context means a model can look at very long documents, codebases or conversations in one go.
- Agents that keep going. Alibaba is aiming the model at multi-step work: planning, using tools and checking results over hours, not just single answers.
- More choice for local AI. If the smaller open-weight Qwen4 models arrive as promised, they could run in local AI apps, the same space that tools like llama.cpp serve.
- Benchmarks are a starting point. Most of the numbers above come from Alibaba itself. The best test is your own tasks.
Quick checklist
- Developers can try
qwen4-maxon QwenCloud through the API or Token Plan. - Check where your data is processed before you send private documents to any cloud model.
- Watch the Qwen channels for the open-weight Qwen4 release; no date has been given yet.
Related reading
- llama.cpp 1.0 is out: a stable engine for local AI apps
- Claude Fable 5.2 is out, up to 30% faster than 5.1
- Why AI makes things up: hallucinations explained without jargon
Sources
- Alibaba Cloud Community (Nov 16, 2026): Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery
- QwenCloud: Qwen4-Max model page
- GIGAZINE (Nov 16, 2026): Qwen4-Max unveiled, beating Claude Opus 5.5 in some tests
- Techmeme (Nov 15, 2026, US time): Alibaba releases Qwen4-Max
Read on November 16, 2026.
Sources
- Alibaba Cloud Community (Nov 16, 2026): Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery
- QwenCloud: Qwen4-Max model page
- GIGAZINE (Nov 16, 2026): Qwen4-Max unveiled, beating Claude Opus 5.5 in some tests
- Techmeme: Alibaba releases Qwen4-Max
- Cover image: Alibaba group Headquarters, by Thomas LOMBARD, Wikimedia Commons, CC BY-SA 3.0
Written by our editorial team with the help of AI tools. Spotted a mistake? Tell us in the discussion.