Qwen4 Max’s Nine-Point Gain Comes With A Lower Price And A Remaining Gap
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Qwen4 Max’s Nine-Point Gain Comes With A Lower Price And A Remaining Gap on ThorstenMeyerAI.com

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.

TL;DR

Artificial Analysis scored Alibaba’s Qwen4 Max at 43 on its Intelligence Index, nine points above the previous Qwen flagship. It ranked ninth among 10 models on its listing, while the cost of an index task was reported at $3.27, down from $10.02 for its predecessor. The model is available through Alibaba services, but its weights are closed and launch-day speed data was unavailable.

Alibaba introduced Qwen4 Max on November 16, and independent benchmark company Artificial Analysis published a score of 43 on its Intelligence Index, nine points above the previous Qwen flagship. The same assessment lists Qwen4 Max ninth among 10 models on its model page, while putting the cost of an index task at $3.27, compared with $10.02 for its predecessor.

Artificial Analysis says Qwen4 Max scored above the comparable-model median of 41. The index combines nine evaluations, including agentic coding, office automation, knowledge and physics tasks. The company tested the model at its highest reasoning setting, xhigh; Alibaba’s API default is reported as high, so results from a default configuration may differ. The published ranking is specific to the 10 models shown on Qwen4 Max’s Artificial Analysis page, not a claim about every available AI model.

Artificial Analysis reports a cost of $3.27 per index task for Qwen4 Max, against $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. Costs vary with usage and task, so the benchmark-task figure should not be read as a fixed price for every request.

Alibaba separately reports improvement on its in-house Work360 benchmark: Qwen4 Max scored 52.69, versus 47.74 for the prior model. The company says average cost per Work360 task fell from $4.43 to $1.67. Those figures are vendor-reported results from Alibaba’s own test, rather than an independent evaluation. The company also describes coding, manufacturing and other demonstrations, but the supplied material does not establish them as controlled comparisons.

At a glance
reportWhen: Published November 16, 2026; benchmark…
The developmentArtificial Analysis published its first Qwen4 Max benchmark results on November 16, reporting a nine-point improvement over the previous flagship and lower cost per index task, alongside a ninth-place ranking among 10 models on the listing.

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 Benchmark Costs, Limited Ranking

The launch pairs a measurable generational gain with lower reported benchmark-task costs, giving businesses a reason to test Qwen4 Max for workloads where price and capability both matter. The independent results are more informative than launch demonstrations because Artificial Analysis evaluated the model using a common suite, although the score remains one benchmark measure rather than a guarantee of performance on a particular company’s work.

The ranking also sets a limit on the launch-day case for treating Qwen4 Max as a leader. Its score of 43 trails the page’s top score of 65 and several listed models from major providers. Its ninth-place position and above-median score can both be true: the first describes its place in that specific group, while the second compares it with the stated median for similar-price models. Buyers weighing a switch should also account for the model’s closed weights and the absence of speed data.

Products Worth Considering

Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite

Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite

Bose Smart Ultra Dolby Atmos Soundbar with Voice Control, HDMI eARC, Black

Bose Smart Ultra Dolby Atmos Soundbar with Voice Control, HDMI eARC, Black

Jackery Solar Generator 1000 v2 with 200W Solar Panel, 1024Wh LiFePO4 Battery

Jackery Solar Generator 1000 v2 with 200W Solar Panel, 1024Wh LiFePO4 Battery

As an affiliate, we earn on qualifying purchases.

A Closed Model in Alibaba’s Qwen4 Family

Qwen4 Max is the first model in Alibaba’s Qwen4 family. Alibaba says it has a 2.4-trillion-parameter backbone, with 71 billion parameters activated per token, plus 328 billion N-gram embedding parameters. The company describes a sparse mixture-of-experts design, meaning only part of the model is used for each token. It also attributes long-input handling to a combination of Gated DeltaNet layers and Qwen Sparse Attention. These architecture details come from Alibaba; without released weights, outside researchers cannot independently inspect the implementation.

Alibaba says the model handles text, images and video, with reasoning enabled by default. It is offered through QwenCloud’s API, a Token Plan subscription, and the company’s QwenWork, Qoder and QwenStudio apps. The weights are not downloadable. Alibaba says smaller Qwen4 models will be open-sourced, but the source material gives no release date. The Next Web reported that a limited preview began on November 11, before the public launch announcement.

Alibaba’s launch material also gives selected results from its own tests. For example, it says the share of functions passing both functionality and security checks rose from 22.0% on Qwen3.8 Max to 31.8% on Qwen4 Max. The company’s benchmark chart was published as an image, and the supplied source does not independently verify its displayed comparisons. These claims should be kept separate from the Artificial Analysis score.

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

— Alibaba

Speed, Weights and Real-World Results

Artificial Analysis had no speed or latency data for Qwen4 Max on launch day, so users cannot yet compare its response times with those of other models from the supplied benchmark results. The source also does not provide enough detail to independently reproduce Alibaba’s Work360 findings or its demonstrations. How the model performs on specific workloads, at Alibaba’s default high reasoning setting, remains a separate question from its xhigh benchmark score.

The model’s weights remain closed, preventing outside inspection of its architecture and independent reproduction of the company’s training claims. Alibaba has said smaller Qwen4 models will be open-sourced, but has not given a date in the material provided. It is also unclear when additional independent benchmarks, broader user evaluations or comparable latency measurements will become available.

Further Tests Will Clarify Trade-Offs

The next useful evidence will come from additional independent evaluations that test Qwen4 Max on a wider range of tasks and report both capability and speed. Results at Alibaba’s default high reasoning setting would help users compare the model with the xhigh configuration used by Artificial Analysis. Real-world deployment costs will also depend on prompt length, output volume, reasoning effort and the prices customers actually pay.

Alibaba’s stated plan to open-source smaller Qwen4 models may provide a later route for developers who need downloadable weights, but no release date is confirmed in the source material. For now, Qwen4 Max is available through Alibaba’s services, with its independent score, reported task costs and closed-weight status as the clearest launch-day facts. Whether those trade-offs suit a particular use case will require testing against that workload.

Key Questions

What score did Qwen4 Max receive?

Artificial Analysis gave it 43 on its Intelligence Index, nine points above Qwen3.8 Max (0902), which scored 34. The test used Qwen4 Max at the xhigh reasoning setting.

Does ninth place mean Qwen4 Max is below average?

Not on the comparison described by Artificial Analysis: it says the score of 43 is above the comparable-model median of 41. Ninth place refers to the 10 models on Qwen4 Max’s specific model page, where the top score is 65.

How much does an index task cost?

Artificial Analysis reports $3.27 per task for Qwen4 Max, compared with $10.02 for Qwen3.8 Max. Alibaba lists API prices of $1.50 per million input tokens and $4.50 per million output tokens; actual charges depend on token use and service terms.

Can developers download Qwen4 Max?

No. The source says Qwen4 Max has closed weights and is available through Alibaba’s API, subscription and apps. Alibaba says smaller Qwen4 models will be open-sourced, but has not provided a release date in the material supplied.

Are speed and latency known?

Not from the launch-day Artificial Analysis results. The provider had no speed or latency data for Qwen4 Max when it published its first evaluation.

Source: ThorstenMeyerAI.com

You May Also Like

Qblox Integrates NVIDIA DGX Spark With Quantum System Controller

Qblox reports sub-7.2-microsecond round-trip latency integrating NVIDIA DGX Spark with its quantum controller through NVQLink.

How We Plan To Put AI To Work For Australia

OpenAI says it plans Australian research partnerships, expanded tool access and AI education programs, but has not disclosed funding or timelines.

Explore Promptable, Adaptable Transcription With NVIDIA Nemotron 3.5 And Granary V2

NVIDIA has published an open-weight speech model, multilingual speech data and tools for prompting and fine-tuning, alongside company-reported benchmark results.

A Look At Anthropic’s Claude-Inspired Approach To Science

The available material gives only Anthropic’s “Claude-shaped science” headline, not the article details needed to verify a scientific development.