VitoMind · Content Pack

Qwen 4 Max, Tested: Alibaba Promised 60% Cheaper. I Checked the Real Bill.

Client: VitoMind  ·  November 17, 2026

Episode summary

Alibaba released Qwen 4 Max, a 2.4-trillion-parameter mixture-of-experts model with a one-million-token context window, and claimed it does the same work as Qwen 3.8 Max for about 60% less money. VitoMind walks through the launch, the showcases and the independent numbers, then gives both Qwen models the same six build prompts already used on Claude Opus 5.5, GPT-6.1 Sol and three Gemini models. Qwen 4 Max ran six builds out of six against two for its predecessor, and cost 57% less on the real bill, second cheapest of all the models tested.

01Five Shorts

Vertical clips, cut at sentence boundaries. End time is the start of the next sentence.

Short 140 s
Cut
08:05 – 08:45 (40 seconds)
On-screen hook
Same prompt. Same company. One works.
Title
Qwen 4 Max vs Qwen 3.8 Max: the castle test
Why it works
A clear visual contrast in seconds: a bright castle that flies at night next to one stuck on a loading screen.
Short 250 s
Cut
10:16 – 11:06 (50 seconds)
On-screen hook
Alibaba promised 60% cheaper. Here's my bill.
Title
I checked Alibaba's 60% cheaper claim with my own money
Why it works
A promise, a test and a verdict in under a minute, ending with a price ranking of six models people know.
Short 333 s
Cut
06:28 – 07:01 (33 seconds)
On-screen hook
Price per token is only half the bill.
Title
Why cheap AI tokens can still cost you more
Why it works
The taxi meter analogy teaches something useful about every model, not only Qwen, so it travels beyond the launch.
Short 439 s
Cut
04:40 – 05:19 (39 seconds)
On-screen hook
Every AI agent should copy this feature.
Title
Live Delivery: stop losing a whole day to an AI agent
Why it works
It names a pain every agent user knows (waiting hours for a wrong result) and shows a concrete fix.
Short 534 s
Cut
04:06 – 04:40 (34 seconds)
On-screen hook
48 hours alone. One new molecule.
Title
An AI ran a robotic lab for 48 hours. Here's what it made.
Why it works
Strong numbers and a physical result, plus an honest caveat that these are Alibaba's own showcases.

02Newsletter / Blog Article

About 600 words

Qwen 4 Max, Tested: Alibaba Promised 60% Cheaper. On Six Real Builds, It Was 57%.

What Alibaba released

Alibaba has opened Qwen 4 Max to everyone, the flagship and first model of the Qwen 4 family. After a limited preview that started on November 11, it is now available on Qwen Cloud through the API or a token plan, and inside Alibaba's own apps: Qwen Work, Qwen Studio and the coding tool Qoder.

It is a mixture-of-experts model with 2.4 trillion parameters, of which about 71 billion are working at any moment. It reads images and video, and its context window is one million tokens, roughly 1,500 printed pages in one prompt.

The price claim

The price is $1.50 per million tokens in and $4.50 out, down from $2 and $6 for Qwen 3.8 Max. Alibaba's headline promise is the same work for about 60% less money. On its own Work360 test, the average cost per task drops from $4.43 to $1.67, which is 62% less, because the model is cheaper per token and also writes fewer tokens: about a quarter fewer on its highest setting and a third fewer on high.

The showcases

Alibaba's demos are ambitious: 3D scenes built in Blender, games in Unity, Unreal and Godot, production lines and car parts planned with two car-industry partners, a short-film pipeline with WonderClip, an 86-page due diligence report from 198 messy files, and treatment steps proposed from a full intensive-care record. In science, it worked alone for 48 hours, screened 2.7 million candidate molecules and directed a robotic lab to produce a new coolant molecule. These are showcases picked by Alibaba, and Alibaba itself says a human still makes the call in medicine.

The most useful feature may be Live Delivery: instead of waiting hours for an agent to finish, you get a usable first version early, then improvements in steps, and you can stop it or change direction at any point. Alibaba is releasing that plugin as open source. The Max weights are not public, but smaller Qwen 4 models are coming as open source, with no date yet.

The independent numbers

Artificial Analysis puts Qwen 4 Max at 43 on its intelligence index, above the average of its class and three points behind Kimi K3.1 Pro. Claude Opus 5.5 and GPT-6.1 Sol score higher. Its cost per task there is $3.27, second cheapest in the chart after Sol and about two thirds less than Qwen 3.8 Max. One warning: it wrote 127 million tokens to finish that test, against about 79 million for a typical model. It is a talker, and that matters, because price per token is only half the bill. The other half is how many tokens the model decides to write, including thinking you never see.

Six builds, one try each

So I ran my own bill. Qwen 4 Max and Qwen 3.8 Max got the same six prompts already used on Opus 5.5, Sol and three Gemini models, word for word, default thinking setting, no second chances: a life animation, a voxel castle, a short film, a 3D car showroom, a driving game and a music studio.

Qwen 4 Max ran six out of six, with one visual flaw: dark holes in the lower body of the car. Qwen 3.8 Max ran two out of six. Its castle never passed the loading screen because of one syntax error, its film was mostly black from a shader that didn't compile, its car never appeared, and its driving game stopped on the title screen.

The bill

For the six builds, Qwen 3.8 Max wrote 341,000 tokens and cost $2.05. Qwen 4 Max wrote 196,000 tokens and cost $0.89: 57% less, close to the promised 60, and done in about an hour instead of more than two. Across all models on the same prompts, Sol is still cheapest at $0.77, then Qwen 4 Max at $0.89, Gemini 4 Argon at $1.52, Hydrogen at $1.75, Opus 5.5 at $2.70 and Gemini 4 Carbon at $5.21.

Verdict

Qwen 4 Max is better, faster and cheaper than its predecessor at the same time, which rarely happens in one release. It is not the smartest model, and it is not the cheapest. But six working builds out of six for about a third of what Opus cost is high-level work. One sample per build, so treat it as a comparison you can click through, not a benchmark. Every build is on the site.

03Five LinkedIn Posts

Angle: verifying vendor claims

Alibaba said Qwen 4 Max does the same work as Qwen 3.8 Max for about 60% less money.

I don't take launch claims at face value, so I tested it. Same six build prompts for both models, word for word, one try each, and I kept the real bill.

Qwen 3.8 Max: 341,000 tokens, $2.05.
Qwen 4 Max: 196,000 tokens, $0.89.

That's 57% less. Alibaba promised about 60. The promise held, and the new model finished in about an hour instead of more than two.

The lesson for anyone buying AI at work: run your own tasks before you believe any chart, including the vendor's. Sometimes the claim holds. You only know if you check.

Angle: the token meter

Price per token is not the price of a task.

Think of a taxi meter. The vendor advertises the price per click. How many times the meter clicks is up to the model, including the thinking it does before it answers, which you never see.

Example: on Artificial Analysis's test, Qwen 4 Max wrote 127 million tokens, when a typical model writes about 79 million. It's a talker. It still came out cheap per task, because the per-token price is low.

On my own six builds, about 100,000 of its tokens were invisible thinking. Cheap thinking, but thinking you pay for.

If you compare models for your team, compare cost per finished task, not cost per token.

Angle: agent design

The feature I want every AI tool to copy is not a benchmark score.

It's called Live Delivery, and it ships with Alibaba's Qwen 4 Max.

The usual problem with agents: they work for hours, you wait until the end, and if something went wrong at the start, you've lost the day.

With Live Delivery, the model hands you a usable first version early, then improves it in steps. You can stop it and change direction at any point.

Alibaba is releasing the plugin as open source.

That is how people actually work with colleagues: show me a draft, then iterate. Long-running agents should behave the same way.

Angle: one generation apart

Same prompt. Same company. One generation apart.

I gave Qwen 4 Max and Qwen 3.8 Max six identical build tasks: an animation, a 3D castle, a short film, a car showroom, a driving game and a music studio.

Qwen 4 Max: six out of six running, one visual flaw on the car.
Qwen 3.8 Max: two out of six.

The failures were small things: one syntax error, a shader that didn't compile, a variable used before it existed. Each one was enough to make the whole build useless.

That's the real gap between model generations at work. Not a few points on an index, but whether the first try runs.

Angle: reading showcases honestly

How to read an AI launch showcase.

Alibaba's Qwen 4 Max demos are impressive: a due diligence report of 86 pages from 198 messy files, treatment steps from a full intensive-care record, 48 hours of solo work directing a robotic lab to a new coolant molecule.

Two things are true at once.

They show where these models are going.
They are picked by the vendor.

Alibaba itself says a human still makes the call in medicine, and that's right.

So treat showcases as direction, then look for independent numbers and your own tests. For Qwen 4 Max, both held up on cost. On intelligence, Opus and Sol still score higher. Both facts matter.

04X Thread (8 posts)

Each post under 270 characters

1/8 Alibaba says Qwen 4 Max does the same work as Qwen 3.8 Max for about 60% less. I gave both the same six build prompts, one try each, and kept the real bill. Here's what happened.

2/8 The specs: 2.4 trillion parameters, mixture of experts with about 71 billion active at any moment, images and video in, and a 1 million token context. Roughly 1,500 printed pages in one prompt.

3/8 Price: $1.50 per million tokens in, $4.50 out. Qwen 3.8 Max was $2 and $6. On Alibaba's own Work360 test, average cost per task drops from $4.43 to $1.67.

4/8 Independent check: Artificial Analysis gives it 43 on its intelligence index and $3.27 per task, second cheapest after GPT-6.1 Sol. But it wrote 127M tokens vs a typical 79M. It's a talker.

5/8 My test: life animation, voxel castle, short film, car showroom, driving game, music studio. Qwen 4 Max: 6 of 6 running, one visual flaw on the car. Qwen 3.8 Max: 2 of 6.

6/8 The bill. Qwen 3.8 Max: 341K tokens, $2.05. Qwen 4 Max: 196K tokens, $0.89. That's 57% less. Alibaba promised about 60. The promise held, and it took about an hour instead of two-plus.

7/8 Same six prompts, every model: GPT-6.1 Sol $0.77, Qwen 4 Max $0.89, Gemini 4 Argon $1.52, Hydrogen $1.75, Claude Opus 5.5 $2.70, Gemini 4 Carbon $5.21.

8/8 Verdict: better, faster and cheaper than its predecessor at once. Not the smartest (Opus and Sol score higher). One sample per build, so a comparison, not a benchmark. Would you switch?

05YouTube Metadata

Title options

1. Qwen 4 Max Tested: Alibaba Promised 60% Cheaper. I Checked the Bill.

2. Qwen 4 Max vs Qwen 3.8 Max: 6 Builds, Same Prompts, Real Cost

3. Qwen 4 Max Built 6 Apps for 89 Cents. Opus Cost $2.70.

Description

Alibaba just released Qwen 4 Max: 2.4 trillion parameters, a one-million-token context window, and a promise of the same work as Qwen 3.8 Max for about 60% less money. I didn't just read the announcement.

I gave Qwen 4 Max and Qwen 3.8 Max the exact same six build prompts that Claude Opus 5.5, GPT-6.1 Sol and three Gemini models already made for this channel: a life animation, a voxel castle, a short film, a 3D car showroom, a driving game and a music studio. One try each, default settings, and I kept the real bill.

You'll also see the launch, Alibaba's showcases, the Live Delivery feature, the Artificial Analysis numbers, and why price per token is only half the story.

Every build is on my site, including all six from Qwen 3.8 Max exactly as they came back. The exact prompts and full code are for members of the Ruocco Academy (waiting list on the page).

Would you switch to Qwen 4 Max? Tell me in the comments.

Chapters

00:00 Qwen 4 Max is out
00:39 The launch and where to use it
01:03 Mixture of experts and 1M context
01:27 Price and the 60% claim
02:03 Demos: Blender, games, factories, film
03:25 Finance, law and intensive care
04:06 Science: a new coolant molecule
04:40 Live Delivery
05:19 Safety and open source
05:55 Independent numbers
06:37 What a token really costs
07:12 The test: six builds, same prompts
07:44 Build 1: life animation
08:05 Build 2: the castle
08:45 Build 3: the short film
09:15 Build 4: the car showroom
09:36 Build 5: the driving game
09:49 Build 6: the music studio
10:16 The bill
10:45 Against every other model
11:13 Verdict

064-Week Posting Calendar

Every piece in this pack is used once.

13 pieces over 4 weeks
WeekDayPlatformPiece
Week 1 – Launch
1MondayYouTubeFull episode with title, description and chapters
1TuesdayXX thread (8 posts)
1WednesdayYouTube Shorts / TikTok / ReelsShort 1 – The castle test
1ThursdayLinkedInPost 1 – Verifying vendor claims
Week 2 – The cost story
2MondayNewsletter / BlogArticle – Qwen 4 Max, Tested
2WednesdayYouTube Shorts / TikTok / ReelsShort 2 – The bill
2ThursdayLinkedInPost 2 – The token meter
Week 3 – How AI works
3TuesdayYouTube Shorts / TikTok / ReelsShort 3 – Price per token is half the bill
3WednesdayLinkedInPost 3 – Agent design
3FridayYouTube Shorts / TikTok / ReelsShort 4 – Live Delivery
Week 4 – The bigger picture
4TuesdayLinkedInPost 4 – One generation apart
4WednesdayYouTube Shorts / TikTok / ReelsShort 5 – 48 hours, one molecule
4ThursdayLinkedInPost 5 – Reading showcases honestly