AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Artificial Analysis gave Alibaba’s Qwen4 Max an Intelligence Index score of 43, nine points above Qwen3.8 Max (0902). The company says the model costs less to run, but independent launch-day task-cost figures show $3.27 versus $10.02, and its ranking remains near the bottom of the ten models on the page.

Alibaba’s Qwen4 Max scored 43 on the Artificial Analysis Intelligence Index, nine points above Qwen3.8 Max (0902), according to results published on November 16, 2026. The independent benchmark page lists an estimated cost of $3.27 per index task, compared with $10.02 for the predecessor, while ranking Qwen4 Max ninth among ten models shown on its page.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

Artificial Analysis’s score places Qwen4 Max above the 41-point median for models in a similar price range, the source material reports. The index runs from 0 to 70; the highest-scoring model on the Qwen4 Max page has 65. The nine-point difference from Qwen3.8 Max is a change in index score, not a claim that the model is nine percent better across all uses.

The benchmark’s estimated cost per task fell from $10.02 to $3.27. The source also gives Qwen4 Max list prices of $1.50 per million input tokens and $4.50 per million output tokens. These figures refer to the index task and listed API pricing; actual bills can vary with usage and task length.

Alibaba reports a separate improvement on its in-house Work360 test: Qwen4 Max scored 52.69, up from 47.74 for Qwen3.8 Max, while average cost per task fell from $4.43 to $1.67. Those are company-reported results on a vendor-defined benchmark, not independent measurements. Artificial Analysis had no speed or latency data for Qwen4 Max on launch day.

At a glance
reportWhen: Published November 16, 2026; launch-day…
The developmentArtificial Analysis published its first Qwen4 Max benchmark results on November 16, 2026, showing a nine-point gain over its predecessor and a lower estimated task cost.
Bar chart titled Ninth of ten on its own page, nine points above its predecessor. Artificial Analysis Intelligence Index for the 10 models on the Qwen4 Max page: Claude Fable 5.2 (max) 65, Gemini 4 Carbon (high) 60, Claude Opus 5.5 (max) 59, GPT-6 Astra (max) 57, GPT-6.1 Sol (max) 55, Muse Wonder Full (max) 50, Kimi K3.1 Pro (open weights) 46, Muse Wonder (max) 44, Qwen4 Max 43 (highlighted), Grok 4.7 (xhigh) 35. Below a dashed line: the predecessor Qwen3.8 Max (0902) 34
Qwen4 Max is ninth of the ten models on its Artificial Analysis page and nine points above its predecessor. Source: Artificial Analysis, model pages for Qwen4 Max and Qwen3.8 Max (0902), retrieved November 16, 2026. Chart: Thorsten Meyer AI.

Lower Cost, Still Behind Leaders

The results suggest a potential trade-off for teams choosing an AI model: a higher benchmark score than Alibaba’s previous flagship alongside a lower estimated cost per benchmark task. That could make Qwen4 Max worth evaluating for workloads where price and task quality both matter. The index result alone does not establish how it performs on a particular company’s tasks, or whether it will be faster or cheaper in day-to-day use.

Its position on the same page adds perspective. A score of 43 is above the reported price-group median, but ninth among ten models listed and well below the page leader’s 65. The launch figures therefore point to an improvement over its predecessor rather than a leading overall score on this comparison.

Cost comparisons also depend on the work being measured. Artificial Analysis’s $3.27 figure is tied to an index task; Alibaba’s $1.67 figure comes from Work360. Because the tests differ, readers should not treat the two amounts as competing estimates of the same workload. Independent testing on real use cases, alongside speed and reliability data, would help show whether the lower benchmark cost carries over to production.

Products Worth Considering

Apple Watch Series 10 GPS 46mm with Silver Aluminum Case and Blue Cloud Sport Loop

Apple Watch Series 10 GPS 46mm with Silver Aluminum Case and Blue Cloud Sport Loop

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

As an affiliate, we earn on qualifying purchases.

A Large Model With Closed Weights

Alibaba introduced Qwen4 Max as the first model in its Qwen4 family. The launch blog describes a 2.4-trillion-parameter backbone, with 71 billion parameters activated per token, plus 328 billion N-gram embedding parameters. The company says the model has a one-million-token context window and can process text, images and video. These specifications come from Alibaba; the model’s unpublished weights prevent outside inspection of its architecture and training claims.

The company says Qwen4 Max is available through the QwenCloud API, a Token Plan subscription, and Alibaba apps including QwenWork, Qoder and QwenStudio. Alibaba says smaller Qwen4 models will be open-sourced, but the launch material provides no release date. The source reports that a limited preview began on November 11, before the public launch announcement.

Alibaba also describes improvements on Work360, token use and coding tests, as well as examples involving manufacturing, legal review and scientific work. These figures and demonstrations are vendor claims. The source material does not provide independent replication of those case studies, and its account says Alibaba’s main comparison chart was published as an image rather than in a form its author used to verify individual values.

“The average cost per task drops from $4.43 to $1.67 — a 62% reduction.”

— Alibaba, on its Work360 evaluation

Speed and Real-World Costs

Speed and latency remain unreported in the launch-day Artificial Analysis results. Without those measurements, the available comparison does not show how quickly Qwen4 Max responds or how it handles sustained workloads.

The benchmark figures also leave open how results translate to individual tasks. Alibaba’s Work360 score and cost are company-reported; Artificial Analysis’s index score and task-cost estimate use a different evaluation. The available source material does not provide enough detail to compare those costs on a shared workload or to confirm the vendor’s demonstrations independently.

It is also unclear when Alibaba will release the smaller Qwen4 models it says will be open-sourced. Qwen4 Max itself is not downloadable, according to the source, so independent evaluation is currently limited to access routes Alibaba provides.

More Independent Testing Needed

The next useful updates would be independent speed and latency measurements, additional results across specific tasks, and tests that reproduce the lower-cost claims on clearly described workloads. Those findings would help organizations compare performance and total usage costs against alternatives.

Alibaba has said that smaller Qwen4 models will be open-sourced, but has not announced a date in the material available for this report. Until those weights or further benchmark results are released, the clearest confirmed comparison remains the launch-day index score: 43 for Qwen4 Max versus 34 for Qwen3.8 Max (0902), with lower estimated index-task cost and no reported speed data.

Key Questions

What does the nine-point improvement measure?

Artificial Analysis scored Qwen4 Max at 43 and Qwen3.8 Max (0902) at 34 on its Intelligence Index. The difference is nine index points; it does not mean a nine-percent gain on every task.

How much does Qwen4 Max cost per task?

Artificial Analysis lists an estimated $3.27 per index task, compared with $10.02 for Qwen3.8 Max. Alibaba separately reports $1.67 per task on its Work360 test. The amounts come from different evaluations and should not be treated as directly comparable.

Is Qwen4 Max among the top models on the index?

It scored above the reported median of 41 for models in a similar price range, but ranked ninth of ten models on the Artificial Analysis page. The top model there scored 65.

Can developers download Qwen4 Max?

No. The source reports that Qwen4 Max’s weights are closed and the model is available through Alibaba’s services. Alibaba says smaller Qwen4 models will be open-sourced but has not given a release date.

How fast is Qwen4 Max?

The launch-day Artificial Analysis results did not include speed or latency data. Its response time and performance under sustained use remain unclear.

Source: Thorsten Meyer AI

You May Also Like

What Did GPT-6 Astra Catch In Claude’s Model Edit?

GPT-6 Astra reviewed a Claude-led model-editing experiment, finding benchmark and weight-conversion faults despite passing tests.

Are AI Agents Getting Too Much Access? How To Delegate More Safely

Recent reports describe agents moving crypto, exposing financial details and appearing inside bank accounts, raising questions about permissions and safeguards.

Talorys – A Self-hosted Personal AI Agent On Cloudflare’s Free Tier

Talorys is an open-source assistant designed for a user’s Cloudflare account, with chat, memory and task tools on the free plan, subject to quotas.

I Put Jev’s New Vision To Work On 30 Images I Made

A 30-image test of TypeSafe’s Jev models found strong product-image matching, but results came from a small, author-run evaluation.