Mid-century poster of a giant fish made of geometric modules rising from the sea at sunset while a small figure on a pier measures it with a ruler

Quick answer

Our Qwen 4 Max review in one line: it is a solid, capable model with real strengths in front-end web design, but the full release does not beat the leading models on independent tests or on value. Alibaba released Qwen 4 Max on 16 November 2026 as the first full model of the Qwen 4 family. It has a 1 million token context window, image and video input, tool calling and structured outputs, and costs $1.50 per million input tokens and $4.50 per million output tokens. Independent scoring puts it 43 on the Artificial Analysis Intelligence Index, behind Kimi K3.1 Pro and well behind GPT-6.1 Sol. WorldofAI, who tested an early preview, found the final version weaker than the preview and slow on long tasks.

Key takeaways

  • It is out now, free to try in Qwen Chat and paid through Alibaba’s QwenCloud API and Token Plan.
  • It is much better than Qwen 3.8 Max: +9 points on the Intelligence Index, cheaper per token and, by Alibaba’s figures, around 60% cheaper per task.
  • It is not a frontier leader. It scores 12 points below GPT-6.1 Sol while costing more than twice as much per task in that test, according to the video.
  • Front-end web design is its standout skill, with a number 6 spot on Code Arena’s web development board.
  • Long builds are slow: one game took about two hours and one simulation about a day of rework in the presenter’s tests.
  • Qwen 4 Max is not open-weight. Smaller open-weight Qwen 4 models, including a 27B version, are promised “soon”.

🛠️ Practical guides

Ready to put this into practice? These step-by-step guides answer two common follow-up questions:

On 8 November we covered the Qwen 4 preview, when early testers were posting impressive coding demos and the Max model’s specifications had only leaked. Now the full Qwen 4 Max is here, with official pricing, benchmark scores and independent tests. This Qwen 4 Max review looks at what changed, what the numbers really say and who should bother switching.

It is based on the launch-day test video from WorldofAI, with specifications and prices checked against Alibaba’s announcement and the QwenCloud model page.

Watch: the source video

What this video covers: WorldofAI’s hands-on Qwen 4 Max test: the presenter compares the final model with the early checkpoint he previewed, walks through Alibaba’s claims and independent scores from Artificial Analysis, Code Arena and his own WoAI Bench, shows pricing and access, and reviews front-end, macOS clone, game, simulation and short film demos before giving a frank verdict. The video is sponsored by Abacus AI SuperComputer.

Video: WorldofAI (@intheworldofai). Sponsored by Abacus AI SuperComputer. This summary and the opinions in it are our own. Scores quoted from the video come from the presenter’s screen and may change as benchmarks are updated.

What is Qwen 4 Max?

Qwen 4 Max is Alibaba’s new flagship AI model and the largest member of the Qwen 4 family. Alibaba pitches it as “a teammate for work, creativity and discovery”, built for coding, office tasks, web development and longer, multi-step agent workflows.

FeatureQwen 4 Max
Release date16 November 2026
InputsText, images and video (outputs text)
Context window1 million tokens
ArchitectureMixture of Experts: 2.4 trillion total parameters, about 71 billion active per token (Alibaba)
ToolsWeb search, tool calling and structured outputs
ReasoningAdjustable effort levels, or reasoning switched off
Open-weight?No. Smaller open-weight Qwen 4 models are promised later
Price (international API)$1.50 input / $4.50 output per million tokens

“Mixture of Experts” simply means only part of the model works on each word it produces. That is how a very large model can stay relatively affordable to run.

What is new in the full Qwen 4 Max release?

Compared with what we knew from the preview, the full release confirms several leaked details and adds a few new ones:

  • The 1 million token context window is confirmed. In the preview it only appeared on a deleted partner image.
  • Official pricing is lower than Qwen 3.8 Max. Alibaba’s price list shows Qwen 3.8 Max at $2 input and $6 output per million tokens, so Qwen 4 Max is about 25% cheaper per token.
  • Better token efficiency. Alibaba says it uses about 35% fewer tokens than Qwen 3.8 Max at high reasoning effort on its own Work360 benchmark, and cuts average cost per task by around 60% (the blog chart shows $4.61 falling to $1.67).
  • A new architecture, with about 25% fewer active parameters than Qwen 3.8 Max at the same overall size.
  • It is not open-weight. Many people expected Qwen 4 to follow earlier Qwen models. Only the smaller models are promised as open-weight releases.

🔎 Unconfirmed / rumour: The presenter expects the open-weight Qwen 4 27B model in “maybe two or three weeks”. That is his guess, not an Alibaba date. If you are planning to run it on your own machine, our guide to how much memory you need for an open-weight model will help you prepare.

How does Qwen 4 Max score on independent benchmarks?

Short answer: much better than its predecessor, but mid-pack against the current leaders. Here is what the video shows from independent sources:

TestQwen 4 Max resultContext
Artificial Analysis Intelligence Index43+9 over Qwen 3.8 Max; 3 points behind Kimi K3.1 Pro; 12 points behind GPT-6.1 Sol
Cost to run the indexAbout $3.21 per taskNearly 3x cheaper than Qwen 3.8 Max, about 11% cheaper than Kimi, but more than double GPT-6.1 Sol
Output tokens39% fewer than Qwen 3.8 MaxSupports Alibaba’s efficiency claim
Terminal Bench 4.0 (coding agents)48%Kimi K3.1 Pro scores 67%
Code Arena web developmentNumber 6Its strongest independent result
WoAI Bench (presenter’s own)Behind GPT-6 Sol, GPT-6.1 Terra, Opus 5 and Kimi K3.1 ProOne reviewer’s benchmark, but run on the preview and the release

Alibaba’s own charts tell a rosier story, showing Qwen 4 Max ahead of Claude Opus 5.5, GPT-6.1 Sol and GPT-6 Astra in several areas. The presenter calls this “bench-maxxed”, meaning tuned to look good on specific tests. That is his opinion, but the gap between the company’s charts and independent scores is worth noticing.

💡 Tip: Vendor charts are not lies, but they are chosen. Our practical guide How to read an AI company’s benchmark chart shows you what to check before you trust one.

Why did the full release test worse than the preview?

This was the most surprising part of the video. The presenter had early access to a Qwen 4 checkpoint and was impressed: his preview Minecraft-style build put up “a great fight” against much larger models. Re-running the same tests on the final release, he saw what he describes as a “substantial dip” in quality.

We do not know why. As we noted in our preview coverage, the early access API reportedly routed requests across different Qwen 4 sizes, so the preview demos may not have come from the exact model that shipped. Final models are also often tuned for speed and cost after early testing.

⚠️ Warning: Early access results and launch-day demos are a preview, not a promise. Test the version you will actually pay for, on your own tasks, before you change anything in your business.

How much does Qwen 4 Max cost?

$1.50 per million input tokens and $4.50 per million output tokens on Alibaba’s international API, with cached input at just $0.075 per million. You can also access it through Alibaba’s Token Plan subscription, and try it free in Qwen Chat.

ModelInput / output per million tokensNotes
Qwen 4 Max$1.50 / $4.501M context, thinking tokens billed as output
Qwen 3.8 Max (previous)$2 / $6Alibaba price list
Batch API50% of the aboveFor jobs that can wait

A lower token price does not automatically mean a lower bill. The presenter found Qwen 4 Max slow and token-hungry on big builds, and the Artificial Analysis cost per task is higher than GPT-6.1 Sol’s. Use our guide to comparing AI models on token cost to test it properly, and remember that higher reasoning effort means more output tokens (see how to choose a reasoning effort level).

What is Qwen 4 Max good at in the demos?

Front-end web design is where Qwen 4 Max shines. The presenter says it follows design instructions closely, including typography, scroll animations and a defined “visual taste”, which older Qwen models struggled with.

DemoWhat it showedWatch-out
Landing pages (a software site, a graphics card product page)Polished layouts, strong typography and scroll effects that matched the briefTook longer than rival models to respond
macOS cloneBoot and login screens, dock, Finder, Safari, Photos, Music, Calculator, Stocks and Settings, with coded iconsA demo, not a working operating system
3D showcase gameA playable wizard-cat action game, reportedly built with real engines such as Unity, Unreal and GodotThe presenter says it “feels a little too good to be true”
Dungeon crawler (presenter’s prompt, Qwen Code)Two floors, enemies, chests, levelling, no errorsTook about two hours
GPU simulationClear visual of how tensor cores workSix rounds of fixes and about a day
One-minute animated short filmDrawn and scored in codeImpressive, but niche

The pattern is clear: high quality when it gets there, but slow. If you try it for big builds, set limits first. Our guide How to stop a long AI build task eating your day gives you a simple time, round and token budget.

What can Qwen 4 Max do with agentic office work?

Alibaba’s launch blog highlights professional, multi-step work. These are the company’s own examples, not independent tests:

  • Finance: using 90 MCP tools to research an investment and produce a traceable Excel and PowerPoint report.
  • Law: organising 198 documents into an 86-page due diligence report that flagged 14 legal issues.
  • Healthcare: analysing about 1,400 medical images and patient records to draft recommendations that Alibaba says matched the doctors’ decisions.

These show the direction Alibaba is aiming for: long, tool-heavy tasks inside a large context window. For a small business, the useful lesson is that long tasks need checkpoints and human review, whichever model you use.

What is confirmed and what is only claimed?

ClaimStatus
Released 16 November 2026 with 1M context, image and video inputConfirmed (Alibaba, QwenCloud)
$1.50 / $4.50 per million tokensConfirmed (QwenCloud price list)
2.4T total parameters, 71B activeAlibaba’s figures
~60% lower cost per task than Qwen 3.8 MaxAlibaba’s own benchmark
Beats Opus 5.5 and GPT-6.1 Sol in several areasAlibaba’s claim; independent scores disagree
Intelligence Index 43, $3.21 per taskIndependent (Artificial Analysis, as shown in the video)
Final model weaker than the previewPresenter’s experience
Open-weight 27B in two or three weeksPresenter’s guess

Should you use Qwen 4 Max?

For most people, not as their main model yet. The presenter’s verdict is that he would not use it unless he already had a Qwen plan, because rivals do much the same job “better, cheaper and more efficient”. We think that is fair, with a few exceptions:

  • Worth a try if you already pay for Alibaba’s Token Plan, build lots of landing pages or front-end prototypes, or need a very long context at a moderate price.
  • Probably skip if you mainly need coding agents in the terminal (its Terminal Bench score lags), want the fastest answers, or care most about intelligence per pound.
  • Wait if what you really want is an open-weight Qwen 4 model you can run yourself. That is the 27B release, not Max.

Before switching any model, run the founder litmus test and check the real cost of a task, not just the price list.

How can you try Qwen 4 Max today?

  1. Go to Qwen Chat and sign in. It is free to use.
  2. Pick Qwen 4 Max from the model list and choose a reasoning mode. Start at high, not the maximum.
  3. Paste in one real task, such as a landing page brief or a document summary.
  4. Note how long it takes and how good the result is, then compare with your usual model.
  5. If you want API access, use QwenCloud, set a spending limit and start with a small test budget.

💡 Tip: Do not paste client data or private documents into any new AI tool until you have read its data and privacy terms.

Frequently asked questions

Is Qwen 4 Max free?

You can use it free in Qwen Chat. API use through QwenCloud costs $1.50 per million input tokens and $4.50 per million output tokens.

Is Qwen 4 Max open source?

No. Qwen 4 Max is not open-weight. Alibaba says smaller open-weight Qwen 4 models will be released soon, including a 27 billion parameter version.

Is Qwen 4 Max better than GPT-6.1 Sol or Claude Opus 5.5?

Alibaba’s charts say it leads in some areas, but independent scoring shown in the video puts it 12 points behind GPT-6.1 Sol on the Artificial Analysis Intelligence Index.

Is Qwen 4 Max better than Kimi K3.1 Pro?

On the Intelligence Index it is 3 points behind, and it trails clearly on Terminal Bench 4.0 (48% vs 67%). It is slightly cheaper per task and edges Kimi on one knowledge-work test.

What is Qwen 4 Max best at?

Front-end web design. It ranked sixth on Code Arena’s web development board and followed design briefs closely in the presenter’s tests.

How big is the Qwen 4 Max context window?

1 million tokens, enough to fit several long books or a large codebase in one request.

The bottom line

Qwen 4 Max is a real step up for Alibaba and a strong front-end designer, but it arrives in a crowded field where rivals are smarter, faster or cheaper per task. Try it free, test it on your own work, and keep an eye out for the open-weight Qwen 4 models, which may turn out to be the more useful release for small businesses.

author avatar
Garry Knight
I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

Leave a Reply