🔍 Read the full analysis: Qwen4 Max Delivers A Nine-Point Gain At One-Third The Cost Of Before on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Alibaba launched Qwen4 Max on Nov. 16, 2026, and Artificial Analysis’ first evaluation gave it a score of 43, up from 34 for its predecessor. The independent benchmark’s task cost was $3.27, compared with $10.02 for the previous model; Qwen4 Max still ranked ninth among 10 models on its comparison page, and speed data was unavailable.
Alibaba launched Qwen4 Max on Nov. 16, 2026, and independent evaluator Artificial Analysis scored it 43 on its Intelligence Index, nine points above the previous Qwen flagship. The evaluator put the cost of a benchmark task at $3.27, down from $10.02 for its predecessor, while ranking the new model ninth among 10 models on its Qwen4 Max comparison page.
Artificial Analysis tested Qwen4 Max using its highest reasoning setting, xhigh. Alibaba’s API uses high by default, so the published score may not represent the result users get without changing settings. Artificial Analysis describes the score as above the median of 41 among comparable models, but the model’s page places it well behind the top entries: the highest score shown is 65.
The evaluator’s task-cost estimate was $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 benchmark-task cost and per-token prices describe different measures; actual charges depend on token use and task.
Alibaba also reports gains on its own Work360 professional-task benchmark: Qwen4 Max scored 52.69, versus 47.74 for Qwen3.8 Max. The company says average Work360 task cost fell from $4.43 to $1.67. Those figures are vendor-reported, not independent evaluations, and the benchmark’s results should not be treated as interchangeable with Artificial Analysis’ score or cost estimate.
Qwen4 Max · launch day
From launch claim to independent check
Alibaba’s numbers come first. An outside measurement shows where the model actually lands.
- The launchAlibaba calls Qwen4 Max a 24/7 “teammate”, with 2.4 trillion parameters and a one-million-token context window.
- The vendor’s own testOn its in-house Work360 benchmark, Alibaba reports 52.69, up from 47.74 for Qwen3.8 Max.
- The independent indexArtificial Analysis scores it 43, up from 34 and just above the median of 41 for models at a similar price.
- 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 Not a Top Ranking
The initial results point to a trade-off for organizations comparing AI models: Qwen4 Max appears to deliver a measurable improvement over Alibaba’s previous flagship at a lower cost on the Artificial Analysis task measure, but it does not lead the evaluator’s broader comparison. Buyers weighing the model will need to compare price with performance on their own tasks, not rely on one index score.
The ranking adds perspective to Alibaba’s launch claims. Qwen4 Max scored above the median for comparable models, yet sat ninth among 10 models on the page. The top score was 65, and other listed models—including Claude Fable 5.2, Gemini 4 Carbon and Claude Opus 5.5—scored higher. These results do not establish how the systems compare on every use case, but they show that a gain over a predecessor is not the same as leading the field.
Access also matters: Alibaba says Qwen4 Max is closed-weight and API-only. Businesses can use it through Alibaba’s services, but cannot download the model weights to run or inspect it independently. That limits options for local deployment and external verification.
Popular right now

KitchenAid 5.5-Quart Bowl-Lift Stand Mixer, Porcelain White

De'Longhi La Specialista Arte Evo Semi-Automatic Espresso Machine with Grinder and Cold Brew

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink
As an affiliate, we earn on qualifying purchases.
A Large, Sparse Qwen4 Model
Qwen4 Max is the first model in Alibaba’s Qwen4 family. Alibaba describes it as having a 2.4-trillion-parameter backbone, with 71 billion parameters activated per token and an additional 328 billion N-gram Embedding parameters. The company says this sparse mixture-of-experts design activates only part of the network for each token, rather than using the entire backbone on every step.
Alibaba says the model accepts text, images and video and generates text, and describes changes to its attention layers and training architecture. Those technical descriptions and its claim that training had no loss spikes come from the company; because the weights are not public, outside researchers cannot directly inspect the model or verify those implementation claims.
The launch material also cites Alibaba’s own results, including a secure-coding measure that it says rose from 22.0% to 31.8%. It describes demonstrations in manufacturing, legal review and other fields. These are vendor-reported tests and case studies, not independently controlled evaluations. Artificial Analysis’ index provides a separate comparison, but it does not establish performance across every task or deployment.
“The average cost per task drops from $4.43 to $1.67 — a 62% reduction.”
— Alibaba, on its Work360 evaluation
Speed, Access and Test Limits
Artificial Analysis had no speed or latency data for Qwen4 Max on launch day, so the initial evaluation does not show how quickly it responds or how throughput compares with rivals. Its headline score was measured at xhigh reasoning effort, while Alibaba’s API defaults to high; the size of any score difference between those settings is not provided.
The available results also leave questions about performance outside the tested benchmarks. Alibaba’s Work360 and coding claims are company-reported, and its demonstrations are not described as controlled, independently replicated trials. Artificial Analysis’ index combines nine evaluations, but a composite score cannot establish which model will perform best for a particular business workflow.
Alibaba has not released Qwen4 Max weights. It says smaller Qwen4 models will be open-sourced, but has given no release date in the source material. It is also unclear whether the launch-day API prices will remain unchanged or how costs will vary across different workloads.
Watch for Independent Replication
The next useful evidence will be additional independent tests, especially results that compare Qwen4 Max at the API’s default high setting and publish latency and throughput figures. Replication across coding, office tasks and multimodal inputs could help users judge whether the index improvement holds for the work they need the model to do.
Alibaba has said it plans to open-source smaller Qwen4 models, although it has not announced when. Until weights or further technical details are available, Qwen4 Max remains an API-access model, and external verification of its architecture claims remains limited. Users assessing it now can compare the stated token rates with their expected use and test the model directly on their own tasks.
Key Questions
What did Artificial Analysis score Qwen4 Max?
It scored 43 on the Intelligence Index, nine points above Qwen3.8 Max (0902), which scored 34. The test used the xhigh reasoning setting.
How much cheaper was Qwen4 Max in the independent evaluation?
Artificial Analysis estimated a benchmark-task cost of $3.27, compared with $10.02 for the predecessor. This is a task-cost estimate, not a per-token price.
Does Qwen4 Max lead the models in the comparison?
No. It ranked ninth among 10 models on its Artificial Analysis page. Its score was above the stated median for comparable models, but several listed systems scored higher.
Can users download Qwen4 Max?
No. According to the source material, Qwen4 Max is closed-weight and available through Alibaba’s services. Alibaba says smaller Qwen4 models will be open-sourced but has not provided a date.
What is still unknown about the model?
Launch-day reporting included no independent speed or latency data. The gap between results at xhigh and Alibaba’s default high setting, and performance on users’ specific workloads, also remains unclear.
Source: ThorstenMeyerAI.com
