Qwen4 Max In The AI Race: Better Scores, Lower Cost, Still Trailing Leaders
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Qwen4 Max In The AI Race: Better Scores, Lower Cost, Still Trailing Leaders on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

  • Fast, free delivery for dorm and study essentials
  • Prime Video and Amazon Music included
  • Member-only deals
Try Prime for Young Adults Free trial for eligible 18–24 year olds
As an affiliate, we earn on qualifying purchases.

TL;DR

Alibaba launched Qwen4 Max on Nov. 16, 2026, and Artificial Analysis measured a score of 43 on its Intelligence Index, nine points above the previous Qwen flagship. The benchmark firm also put the cost of an index task at $3.27, down from $10.02 for its predecessor, while ranking Qwen4 Max ninth among 10 models on its comparison page. Alibaba’s own benchmark results are vendor-reported, and speed data and downloadable weights were not available at launch.

Alibaba launched Qwen4 Max on Nov. 16, describing it as a multimodal AI model with a 2.4-trillion-parameter backbone and a one-million-token context window. An initial assessment by benchmarking company Artificial Analysis put it at 43 on the firm’s Intelligence Index—up from 34 for its predecessor—and found a lower estimated cost per index task, but ranked it ninth among 10 models on its Qwen4 Max comparison page.

Artificial Analysis tested Qwen4 Max at its xhigh reasoning setting. The company’s Intelligence Index v5.0.1 combines nine evaluations covering areas such as agentic coding, office automation, knowledge and physics. It described the score of 43 as above the median of 41 among models in a similar price range. The ninth-place ranking refers specifically to the 10 models listed on the Qwen4 Max page; the top score there is 65.

The benchmark firm estimated an index-task run at $3.27 for Qwen4 Max, compared with $10.02 for Qwen3.8 Max (0902). Alibaba’s listed API prices are $1.50 per million input tokens and $4.50 per million output tokens. The task-cost estimate is not the same as a fixed charge for every user request: actual costs depend on token use and the task being run.

Alibaba’s launch post reports separate, in-house results. On its Work360 professional-task benchmark, the company says Qwen4 Max scored 52.69, compared with 47.74 for Qwen3.8 Max, and that average cost per task fell from $4.43 to $1.67. Alibaba also reports improved secure-coding results and lower token use on selected evaluations. These are vendor-reported figures, not independent measurements. Its cross-model benchmark chart was published as an image, and the source report did not reproduce values from that graphic.

At a glance
reportWhen: Announced and independently benchmarked…
The developmentAlibaba released Qwen4 Max, while an independent benchmark found improved scores and lower task costs than its predecessor but a ranking near the bottom of the leading-model comparison.

Qwen4 Max · launch day

From launch claim to independent check

Alibaba’s numbers come first. An outside measurement shows where the model actually lands.

Alibaba: 2.4T parameters, 71B active Own test Work360: 47.74 → 52.69 Independent index (0 to 70) 0 70 34 41 43 65 Speed: no data yet
  1. The launchAlibaba calls Qwen4 Max a 24/7 “teammate”, with 2.4 trillion parameters and a one-million-token context window.
  2. The vendor’s own testOn its in-house Work360 benchmark, Alibaba reports 52.69, up from 47.74 for Qwen3.8 Max.
  3. The independent indexArtificial Analysis scores it 43, up from 34 and just above the median of 41 for models at a similar price.
  4. What is still openThe top model on the same page scores 65. Speed is not measured yet, and the weights are closed.

Sources: Alibaba/Qwen launch blog, Nov 16, 2026 (vendor figures); Artificial Analysis, model pages for Qwen4 Max and Qwen3.8 Max (0902), retrieved Nov 16, 2026. Schematic scale; marker positions follow the index values.

Graphic: Powered by Thorsten Meyer AI

Lower Costs, But a Ninth-Place Score

The launch gives buyers evidence of a measurable generational improvement on one independent index, alongside a substantial reduction in the benchmark firm’s estimated task cost. That combination could matter to organisations weighing model performance against the expense of repeated or long-running AI tasks. It does not establish that Qwen4 Max will be cheaper or more capable for every workload; the index cost is tied to a particular evaluation, and use costs vary with tokens and settings.

The ranking also tempers the launch’s competitive message. Artificial Analysis placed Qwen4 Max below the leading scores on its page: Claude Fable 5.2 (max) scored 65, Gemini 4 Carbon (high) 60, Claude Opus 5.5 (max) 59, GPT-6 Astra (max) 57 and GPT-6.1 Sol (max) 55. The best-known comparison is not a complete measure of real-world utility, but it shows why a higher score than the previous Qwen model does not by itself establish leadership.

Access is another practical factor. Qwen4 Max is offered through Alibaba’s API and services, but its weights are not downloadable. That limits the ability of outside researchers and customers to inspect, modify or run the model independently. The launch may therefore appeal to API users seeking a lower-cost option, while offering fewer deployment choices than a model released with open weights.

Ninja Air Fryer XL, 6.5 Qt, 6-in-1, MaxCrisp, Dark Grey (AF181)

Ninja Air Fryer XL, 6.5 Qt, 6-in-1, MaxCrisp, Dark Grey (AF181)

Ninja Foodi 7-Qt MegaZone FlexBasket Air Fryer, 6-in-1, Black, DZ071

Ninja Foodi 7-Qt MegaZone FlexBasket Air Fryer, 6-in-1, Black, DZ071

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink

As an affiliate, we earn on qualifying purchases.

A New Model in Alibaba’s Qwen Family

Qwen4 Max is presented as the first model in Alibaba’s Qwen4 family. The company describes a sparse mixture-of-experts design with 2.4 trillion backbone parameters and 71 billion activated per token, plus 328 billion N-gram embedding parameters. In this architecture, only parts of the model are activated for a given token; the headline parameter total does not mean all parameters are used for every operation.

Alibaba says the model accepts text, images and video and generates text. Its launch documentation describes changes to attention and residual connections, and says reasoning is enabled at high effort by default. The company also says the model is available through QwenCloud, a Token Plan subscription, and its QwenWork, Qoder and QwenStudio apps. The Next Web reported that a limited preview began Nov. 11.

For comparison, Artificial Analysis lists Qwen3.8 Max (0902), released Sept. 2, at 34 on the same Intelligence Index. Qwen4 Max’s 43 therefore marks a nine-point increase in that test. The firm’s comparison also lists Kimi K3.1 Pro at 46, the strongest Chinese model identified in the source report’s snapshot; that model has open weights, unlike Qwen4 Max.

“In our Work360 evaluation, the average cost per task drops from $4.43 to $1.67 — a 62% reduction.”

— Alibaba

Speed, Access and Results Still Open

Artificial Analysis had no speed or latency data for Qwen4 Max on launch day, so the available assessment does not show how quickly it responds or how that compares with rival models. Its index result used the xhigh reasoning setting, while Alibaba’s API default is high; scores from the default configuration may differ.

Alibaba’s claims about its Work360 results, coding tests and customer examples have not been independently verified in the material available for this report. The source account also notes that its main comparison chart was an image and did not transcribe its data. The Next Web, citing FourWeekMBA, reported results of 75.7 on DeepSWE v1.1 and 48.5 on Terminal-Bench 4.0, but those figures are not the Artificial Analysis Intelligence Index score and should be treated as separately reported benchmark results.

Alibaba has not provided a release date for the smaller Qwen4 models it says it plans to open-source. Qwen4 Max itself remains closed-weight, and outside readers cannot verify the architecture or the company’s account of its training run by inspecting published model weights.

Further Benchmarks and Open-Model Plans

The next useful evidence will be additional independent evaluations, including results that make model settings, prompts and task costs clear enough for comparison. Speed and latency measurements would help buyers judge whether the lower estimated task cost comes with different response times. Results using the API’s default high reasoning setting would also clarify how the launch-day xhigh score relates to ordinary use.

Readers can also watch for updates to the Artificial Analysis model pages and for Alibaba to publish more detail on access, pricing and its planned smaller Qwen4 releases. Until those details arrive, the confirmed picture is limited: Qwen4 Max improves on its predecessor in the initial independent index and costs less per index task in that firm’s estimate, but it remains behind the top-scoring models on the comparison page.

Key Questions

What score did Qwen4 Max receive?

Artificial Analysis gave Qwen4 Max a score of 43 on its Intelligence Index v5.0.1, using the xhigh reasoning setting. The firm reported a median of 41 among comparable models.

Did Qwen4 Max outperform its predecessor?

Yes, on that index: Qwen4 Max scored 43, compared with 34 for Qwen3.8 Max (0902). This result does not establish superiority on every task or benchmark.

Is Qwen4 Max cheaper to run?

Artificial Analysis estimated a cost of $3.27 per index task, compared with $10.02 for the predecessor. Alibaba separately reports lower average costs on its own Work360 benchmark. Actual API bills depend on token use and the task.

Can users download Qwen4 Max?

No. Qwen4 Max is closed-weight and is available through Alibaba’s API and services. Alibaba says smaller Qwen4 models will be open-sourced but has not announced a date.

How does it rank against leading models?

It ranked ninth among 10 models on its Artificial Analysis comparison page. The top score there was 65; Qwen4 Max’s score was 43.

Source: ThorstenMeyerAI.com

You May Also Like

AI Sovereignty Explained: More Than National Labels

An analysis of European AI sovereignty shifting from ‘EU-incorporated’ to ‘not US-incorporated,’ and what this means for global data and AI governance.

What Virginia’s AI Cyber Plan Means Amid Washington’s Call For Less Regulation

Virginia’s Oct. 23 directive orders a statewide cyber plan and AI-company briefings as federal and state governments dispute AI regulation.

Why Gimbals Still Matter in the Age of Digital Stabilization

Just when you thought digital stabilization was enough, discover why gimbals still matter for professional, smooth filmmaking in any environment.

Exploring The Downsides Of AI Quantization To Four Bits

Analysis of the performance decline and risks associated with reducing AI model precision to four bits, highlighting potential pitfalls and implications.