🔍 Read the full analysis: Qwen4 Max Is Cheaper And Better Scoring, Though Leaders Remain In Front on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Alibaba launched Qwen4 Max on Nov. 16, 2026, and Artificial Analysis’s first independent results put it at 43 on the Intelligence Index, nine points above its predecessor. The measured cost per index task was $3.27, compared with $10.02 for the earlier model, but Qwen4 Max ranked ninth among 10 models on its comparison page. Speed data and independently verifiable evidence for Alibaba’s own benchmark claims were not available at launch.
Alibaba launched Qwen4 Max on Nov. 16, and first results from benchmarking firm Artificial Analysis show the model scoring 43 on its Intelligence Index, up from 34 for its predecessor, while costing $3.27 per index task compared with $10.02. The improvement has not put it among the leading models on that index: Qwen4 Max ranked ninth of 10 models listed on its comparison page, with the highest score at 65.
Artificial Analysis said Qwen4 Max scored above the 41-point median among comparable models. Its index combines nine evaluations, including agentic coding, office automation, knowledge and physics. The firm tested the model at its highest reasoning setting, xhigh; Alibaba’s API uses high reasoning effort by default, so results under ordinary settings may differ. Artificial Analysis had no speed or latency data for the model on launch day.
The benchmark provider’s model pages list $3.27 per index task for Qwen4 Max and $10.02 for Qwen3.8 Max (0902). Alibaba’s published API prices are $1.50 per million input tokens and $4.50 per million output tokens. Those are distinct cost measures: the index-task estimate is not the same as a fixed price for every user request, which can vary with token use and task.
Alibaba describes Qwen4 Max as a 2.4-trillion-parameter model with 71 billion parameters activated per token, plus 328 billion N-gram embedding parameters. The company says it can process text, images and video, and has a one-million-token context window. It is available through QwenCloud’s API, a Token Plan subscription and Alibaba apps including QwenWork, Qoder and QwenStudio. The model’s weights are not downloadable.
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 Leaders Stay Ahead
The results give businesses and developers an early, independent point of comparison for a model Alibaba is pitching as an AI teammate. The combination of a higher score than its predecessor and a lower measured task cost could make Qwen4 Max worth evaluating for workloads where price and capability both matter. The cost figures are benchmark estimates, however, not a guarantee of what a particular deployment will cost.
The ranking also sets a limit on the launch claims that outside evidence can currently support. Artificial Analysis placed Qwen4 Max below the leading models 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. Qwen4 Max’s score of 43 is an improvement, not a lead over those listed competitors.
Access is another practical distinction. Because Qwen4 Max is closed-weight and API-based, users cannot download it and run it independently. That may affect organisations weighing control, deployment options and the ability to inspect or adapt a model alongside its benchmark performance and price.
Products Worth Considering

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Liquid Retina Display, and Touch ID (Midnight)

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink

Bose Smart Ultra Dolby Atmos Soundbar
As an affiliate, we earn on qualifying purchases.
From Qwen3.8 to Qwen4 Max
Alibaba presented Qwen4 Max as the first model in its Qwen4 family. In its launch blog, the company says the architecture combines Gated DeltaNet layers with a new “Qwen Sparse Attention” mechanism for long inputs, uses four parallel residual branches, and adds a large lookup table for common token patterns. Alibaba also says training proceeded without loss spikes. Those technical descriptions and training claims come from the company; with the weights unpublished, outsiders cannot directly inspect the model or independently verify those details.
Alibaba’s in-house Work360 evaluation reports a score of 52.69, up from 47.74 for Qwen3.8 Max. The company also says average cost per Work360 task fell from $4.43 to $1.67. These results are not the same as Artificial Analysis’s independent index score or task-cost estimate, and Work360 is Alibaba’s own benchmark. The launch blog also cites lower token use on Work360 and Terminal-Bench 4.0, plus vendor case studies in manufacturing, chemistry, legal review and critical care. Those demonstrations are company-reported examples, not controlled independent tests.
External reports cited a mixed coding picture from benchmark results presented by Alibaba. The Next Web, citing FourWeekMBA, reported a DeepSWE v1.1 score of 75.7 for Qwen4 Max versus 74.2 for Claude Opus 5.5, but a Terminal-Bench 4.0 score of 48.5 versus 59.6 for Opus 5.5 and 67.2 for Kimi K3.1. These figures were attributed to Alibaba’s comparisons, rather than presented as independent measurements. A limited preview reportedly opened on Nov. 11, according to The Next Web.
“Your new teammate has arrived.”
— Alibaba’s Qwen team, in its launch introduction
Evidence Still Missing at Launch
The independent picture is limited to the measurements Artificial Analysis had published on launch day. The firm reported no speed or latency data, so users cannot yet compare response times from this evidence. It is also not clear how Qwen4 Max will score under Alibaba’s default high reasoning setting, because the reported index test used xhigh.
Alibaba’s Work360, coding and case-study claims have not been independently established by the material available here. The model’s closed weights also prevent outside researchers from directly checking its architecture and training claims. Costs in real use will depend on the task, token volume, settings and service arrangement; the benchmark task estimate does not settle those costs.
Alibaba said smaller Qwen4 models would be open-sourced, but did not provide a release date. The launch material does not establish when further independent tests, expanded availability or more performance data will appear.
More Tests and Open-Model Plans
The next useful evidence will be independent testing across more tasks and settings, including speed, latency and results at the API’s default reasoning effort. Comparisons under consistent conditions could clarify how much the improvement depends on the evaluation setting and whether the measured price advantage holds for common workloads.
Users considering the API can compare its published token prices with their expected usage, while treating benchmark task costs and Alibaba’s Work360 figures as separate measures. Alibaba has said smaller models in the Qwen4 family will be open-sourced, but has not announced a timetable. Until further results or a release schedule emerge, Qwen4 Max’s independently measured standing is the clearest available guide: a marked gain over Qwen3.8 Max, but still behind the leaders on the reported index.
Key Questions
What did Qwen4 Max score on the independent benchmark?
Artificial Analysis gave Qwen4 Max a score of 43 on its Intelligence Index. It reported a median of 41 among comparable models and ranked Qwen4 Max ninth among the 10 models on its page.
How much cheaper was Qwen4 Max in the reported comparison?
Artificial Analysis listed a cost of $3.27 per index task, compared with $10.02 for Qwen3.8 Max (0902). This is a benchmark task estimate, not a fixed charge for every real-world request. Alibaba separately reported lower costs on its own Work360 evaluation.
Is Qwen4 Max ahead of leading AI models?
No. Its score of 43 was below the top score of 65 on the Artificial Analysis comparison page. The results show a gain over its predecessor, not a lead over the models ranked highest there.
Can users download Qwen4 Max?
No. Qwen4 Max is closed-weight and offered through Alibaba services, including its API. Alibaba said smaller Qwen4 models would be open-sourced but did not give a release date.
What remains unknown about the model?
Artificial Analysis had not published speed or latency data on launch day. Independent verification of Alibaba’s in-house benchmark and case-study claims was also not available, and results at the API’s default reasoning setting may differ from the reported xhigh test.
Source: ThorstenMeyerAI.com
