AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Qwen4 Max Explained: Better By Nine Points, Cheaper, Still Trailing on ThorstenMeyerAI.com

TL;DR

Alibaba launched Qwen4 Max on Nov. 16, and Artificial Analysis gave it an Intelligence Index score of 43, up from 34 for the previous flagship. The independent evaluator ranked it ninth of 10 models on the Qwen4 Max page; its measured index-task cost was $3.27, compared with $10.02 for its predecessor. Speed data was unavailable, and the model’s weights are closed.

Alibaba launched Qwen4 Max on Nov. 16, and independent evaluator Artificial Analysis scored it 43 on its Intelligence Index—nine points above the previous Qwen flagship. The same evaluation puts Qwen4 Max ninth among 10 models on its comparison page, while measuring a lower cost per index task than for its predecessor.

Artificial Analysis tested Qwen4 Max at its highest reasoning setting, xhigh, and reported a score of 43. Qwen3.8 Max (0902), the prior flagship, scored 34. Artificial Analysis describes the new model as above the median of 41 for comparable models, but the page’s ranking places it near the bottom of that group. The highest score shown there is 65.

The evaluator puts the cost of running an index task 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. These figures describe different measures: the per-task estimate reflects the benchmark task, while the API prices are charged by token. Artificial Analysis had no speed or latency data for the new model on launch day.

Alibaba describes Qwen4 Max as a 2.4-trillion-parameter model, with 71 billion parameters activated per token and an additional 328 billion N-gram Embedding parameters. The company says it supports text, images and video, and that reasoning is on by default at the high-effort setting. The model is available through QwenCloud’s API, a subscription called Token Plan and Alibaba apps including QwenWork, Qoder and QwenStudio.

At a glance
reportWhen: Launched and independently evaluated No…
The developmentArtificial Analysis published its first benchmark results for Alibaba’s newly launched Qwen4 Max, finding a nine-point improvement over its predecessor alongside lower measured task costs, but a ninth-place ranking among 10 models on its comparison page.

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 results put Qwen4 Max in a mixed position for organizations choosing a model. The nine-point gain over its predecessor indicates progress on Artificial Analysis’s combined tests, while the lower task-cost estimate may matter to customers running repeated workloads. But a score of 43 and a ninth-place position among 10 listed models do not establish that it matches the highest-scoring alternatives overall.

The gap between a benchmark score and a production decision matters. Artificial Analysis’s index combines nine evaluations, including agentic coding, office automation, knowledge and physics. A ranking across that mix may not predict results on a particular company’s tasks. Buyers would need to test their own workloads and account for reasoning settings, token use, latency and API pricing. Because no launch-day speed data was available, the cost results alone do not show how quickly the model completes work.

Access is another practical distinction: Qwen4 Max has closed weights and cannot be downloaded, according to the source report. That limits independent inspection and local deployment compared with models whose weights are available. Alibaba says smaller Qwen4 models will be open-sourced, but has not given a release date.

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink

As an affiliate, we earn on qualifying purchases.

Alibaba’s Claims and Independent Tests

Alibaba introduced Qwen4 Max as the first model in its new Qwen4 family, describing it as a “24/7 AI teammate” for work, creativity and discovery. Its launch material highlights a sparse mixture-of-experts design, in which only part of the model is activated for each token, as well as new attention and residual-branch components. The company also says training completed without loss spikes. Those architecture and training statements come from Alibaba; the weights are not public, so outside researchers cannot directly verify them.

Alibaba’s in-house Work360 evaluation reports a score of 52.69 for Qwen4 Max, up from 47.74 for Qwen3.8 Max. The company says average cost per Work360 task fell from $4.43 to $1.67. It also reports improved results on secure coding and selected customer examples, including a manufacturing programming task. These are vendor-reported findings and case studies, not results from Artificial Analysis’s independent index or necessarily controlled comparisons.

Some reports citing Alibaba’s benchmark chart describe a mixed performance across coding tests: The Next Web, citing FourWeekMBA, reported a score of 75.7 on DeepSWE v1.1 and 48.5 on Terminal-Bench 4.0. The same report said Qwen4 Max trailed Claude Opus 5.5 on Terminal-Bench. Since Alibaba’s comparison chart was published as an image, the source report did not independently reconstruct its values. The figures should therefore be treated as reported vendor comparisons, distinct from Artificial Analysis’s published index score.

„“In our Work360 evaluation, the average cost per task drops from $4.43 to $1.67.”“

— Alibaba

Open Weights and Missing Speed Data

Artificial Analysis had no latency or speed measurements for Qwen4 Max at launch, leaving an important part of real-world use unassessed. The source material also does not specify when such data may become available. A lower benchmark-task cost does not, on its own, establish that the model is faster or cheaper across different customer workloads.

Qwen4 Max’s weights are unavailable for download, and Alibaba has not set a date for the smaller Qwen4 models it says it plans to open-source. That means the model’s architecture and training claims cannot be checked directly by outside researchers. Alibaba’s in-house benchmark results and customer examples also have not been independently validated in the material available here.

Benchmark rankings depend on the models included, test versions and reasoning settings. Artificial Analysis tested Qwen4 Max at xhigh, while Alibaba’s API default is high. It is not yet clear how much that difference changes scores in typical use, or how the model performs on individual tasks outside the published evaluations.

Further Tests and Open-Model Timing

The next useful evidence will be additional independent testing, particularly speed and latency results and evaluations on specific coding, office and multimodal workloads. Comparisons using the same reasoning settings could also help explain how Qwen4 Max’s high-effort benchmark result translates to its API default.

Alibaba has said that smaller models in the Qwen4 family will be open-sourced, but has not announced a schedule. Until those models or Qwen4 Max weights become available, outside verification of the company’s architecture claims remains limited. Customers considering the API can compare their own task costs and output quality against alternatives; the launch-day benchmarks do not settle those questions.

Key Questions

What did Artificial Analysis find about Qwen4 Max?

It scored 43 on the Intelligence Index, nine points above Qwen3.8 Max (0902), which scored 34. It ranked ninth among the 10 models listed on the Qwen4 Max comparison page.

Is Qwen4 Max cheaper than its predecessor?

Artificial Analysis estimated a cost of $3.27 per index task, compared with $10.02 for Qwen3.8 Max (0902). Alibaba separately reported lower average costs on its own Work360 benchmark. Actual expenses vary by workload and token use.

Can users download Qwen4 Max?

No. The model is available through Alibaba’s services, but its weights are closed and cannot be downloaded. Alibaba says smaller Qwen4 models will be open-sourced, without announcing a date.

How fast is Qwen4 Max?

Its launch-day evaluation did not include speed or latency data. The available cost figures do not show how quickly it responds.

Are Alibaba’s benchmark and customer claims independently verified?

Alibaba’s Work360 scores, cost figures and customer examples are company-reported. Artificial Analysis provides a separate evaluation, but its index does not independently verify every claim in Alibaba’s launch material.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Efficiently Price Loose Lego Bricks With A Simple Tool

A new app allows Lego collectors to estimate the value of loose brick piles using a simple photo-based tool, streamlining resale decisions.

From Chatbot To Marketplace: Anthropic Adds 2,000+ Plugins To Claude

A BleepingComputer headline describes a Claude marketplace with more than 2,000 plugins and connectors, but launch and access details are unavailable.

Boost SEO Resilience In Ecommerce Migrations With Redirect-Map Insurance

Ecommerce platforms can now test redirect-map insurance to prevent traffic loss during migrations, with early validation tools promising fewer errors pre-launch.

What DeepSeek-V4-Flash-High’s Ninth Point Means For AI Development

Analysis of DeepSeek-V4-Flash-High’s ninth point on the Arena leaderboard reveals post-training improvements, influencing AI capabilities and cost-efficiency.