TL;DR
Alibaba launched Qwen4 Max on November 16, 2026, describing a large model with vision and a one-million-token context window. Artificial Analysis scored it 43, up from 34 for the previous Qwen flagship, and put the cost of an index task at $3.27 versus $10.02; its launch-day page ranked it ninth among 10 listed models.
Alibaba introduced Qwen4 Max on November 16, 2026, and independent benchmarking company Artificial Analysis published its first results that day: the model scored 43 on its Intelligence Index, compared with 34 for the previous Qwen flagship, while the listed cost of an index task fell to $3.27 from $10.02. The results show an improvement over its predecessor, while Artificial Analysis’s page placed Qwen4 Max ninth among 10 models listed there.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Alibaba describes Qwen4 Max as the first model in the Qwen4 family. Its launch blog says the model has 2.4 trillion backbone parameters, 71 billion activated per token and an additional 328 billion N-gram Embedding parameters. The company also lists native vision and a one-million-token context window, and presents the model as a “24/7 AI teammate.” These are company descriptions of the model and its intended uses.
Artificial Analysis’s launch-day comparison gives Qwen4 Max an Intelligence Index score of 43, up from 34 for Qwen3.8 Max (0902). The page listed the median for comparable models at 41 and the highest score among its 10 models at 65. The index task cost was $3.27 for Qwen4 Max and $10.02 for its predecessor. The page also listed API prices of $1.50 per million input tokens and $4.50 per million output tokens.
Alibaba’s own results show a similar direction, but they come from company-run tests. The firm reports that Qwen4 Max scored 52.69 on Work360, compared with 47.74 for Qwen3.8 Max, and that average cost per task on that test fell from $4.43 to $1.67. Alibaba also reports lower token use on Work360 and Terminal-Bench 4.0, and a higher pass rate on a secure-coding benchmark. Those figures have not been independently verified in the material provided.

A Lower Measured Cost Per Task
The results give buyers and developers an early comparison on two practical questions: how the new model performs against Alibaba’s previous flagship, and what a measured task costs. In Artificial Analysis’s index, Qwen4 Max is nine points above its predecessor; its listed task cost is less than the predecessor’s. That makes the launch relevant to teams weighing model performance against API spending.
The ranking also puts the gain in perspective. A score of 43 is above the page’s stated comparable-model median of 41, but the model ranked ninth of 10 on that page, whose highest score was 65. The result therefore does not establish that Qwen4 Max leads the wider field. It is one index and one launch-day snapshot, and the supplied material does not detail how well its score predicts performance on a particular organization’s work.
Alibaba’s Work360 cost figures point to a substantial reduction within its own evaluation, but the test is proprietary. Its score, token-use and cost claims should be read as vendor-reported results until outside evaluators can reproduce them or publish comparable tests. For organizations considering adoption, the independent index and the company’s demonstrations answer different questions and have different evidentiary weight.
Products Worth Considering

Bose Smart Ultra Dolby Atmos Soundbar

Bose QuietComfort Headphones, White Smoke

Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black
As an affiliate, we earn on qualifying purchases.
From Qwen3.8 to Qwen4 Max
Qwen4 Max is the first model Alibaba has announced in the Qwen4 family. Alibaba says its sparse mixture-of-experts architecture activates 71 billion parameters per token from a much larger backbone. The company describes a combination of Gated DeltaNet layers and “Qwen Sparse Attention” for long inputs, along with parallel residual branches and an N-gram Embedding component. These architecture details are from the launch blog; the model’s weights are not publicly available for outside inspection.
Alibaba says the model can read text, images and video and generate text. Its listed access routes include the QwenCloud API, a Token Plan subscription, and the company’s QwenWork, Qoder and QwenStudio apps. The source material says a limited preview began on November 11, citing The Next Web. It does not provide details on availability by region or the terms of that preview.
The launch blog also describes company-selected use cases, including manufacturing, legal review and a critical-care case. Alibaba reports that a manufacturing task involving partner companies took about 24 minutes to program, down from approximately 475 minutes manually. These examples are vendor case studies, rather than controlled, independently assessed comparisons. The launch blog’s multi-benchmark comparison chart was published as an image; the supplied reporting does not reproduce its values.
“Your new teammate has arrived.”
— Alibaba’s Qwen team
Speed and Open Weights
Artificial Analysis had no speed or latency data for Qwen4 Max on launch day. The available figures therefore do not show how quickly it responds or how its throughput compares with other models. Those measures can matter to applications where response time, rather than cost per benchmark task alone, shapes the user experience.
The weights are closed, so outside researchers cannot download the model to inspect or independently reproduce its architecture and behavior. Alibaba says smaller Qwen4 models will be open-sourced, but the supplied material gives no release date. It is also unclear when more independent evaluations will become available or whether they will test the company’s Work360 and case-study results.
The reported scores and costs do not establish how the model will perform on a specific customer’s tasks. Artificial Analysis’s index is a single comparative measure, while Alibaba’s Work360 figures come from its own evaluation. The launch material does not include enough detail to independently assess the task mix, scoring process or conditions behind every reported company result.
More Independent Results Needed
The next useful evidence will be independent measurements of speed, latency and task performance, alongside repeatable comparisons of the costs reported by Alibaba and Artificial Analysis. Such results could help organizations judge whether the benchmark improvement translates to their own workloads. The launch-day figures alone do not answer that question.
Alibaba’s stated plan to open-source smaller Qwen4 models is another milestone to watch, although no date was provided. A public release could allow researchers to inspect those models directly; it would not, by itself, establish that Qwen4 Max’s weights will be released.
For now, Qwen4 Max is available through Alibaba’s named API, subscription and app offerings, according to the launch material. Its early independent result indicates a higher index score and lower task cost than its predecessor in the cited comparison. Its speed, broader ranking and performance outside the published evaluations remain open questions.
Key Questions
What did Artificial Analysis report about Qwen4 Max?
Artificial Analysis gave Qwen4 Max an Intelligence Index score of 43, compared with 34 for Qwen3.8 Max (0902). Its page listed a $3.27 cost for an index task for Qwen4 Max, versus $10.02 for the predecessor.
Does Qwen4 Max lead the models in the comparison?
No. The Artificial Analysis page listed Qwen4 Max ninth among 10 models. The highest score shown there was 65, and the comparable-model median was 41.
What cost reduction did Alibaba report?
Alibaba said its Work360 average cost per task fell from $4.43 to $1.67. That is a vendor-reported result from the company’s own evaluation. It is separate from Artificial Analysis’s index task-cost figures.
Can developers download Qwen4 Max?
No. The source material describes Qwen4 Max as having closed weights and says it is available through Alibaba’s API, subscription and apps. Alibaba said smaller Qwen4 models would be open-sourced but did not give a date.
Is Qwen4 Max’s speed known?
Artificial Analysis had no speed or latency data for the model on launch day, so the supplied results do not establish how quickly it responds.
Source: Thorsten Meyer AI
