Today ainotis Join

My notis

ModelsPublished Top storyAll news from that day

Ongoing case: Frontier model API prices New model · 20 stories

Alibaba releases Qwen4-Max, a multimodal flagship with a 1M-token context window

Alibaba says Qwen4-Max beats Qwen3.8-Max on its own Work360 test and ranked 6th on Code Arena; it costs $1.50 input and $4.50 output per 1M tokens in Singapore, less elsewhere.

Share

Check our sources · 18 facts from 3 sources
Illustration: A white access kiosk with a tall display column and a card slot; the green slot stands for the new flagship model opening through the Token Plan and API.Illustration made with AI for ai notis
source · Illustration made with AI for ai notis; not a picture of what happened

Key points

  1. Alibaba says the mixture-of-experts model has 2.4 trillion total parameters and activates 71 billion per token, 25% fewer activated parameters than Qwen3.8-Max.
  2. Alibaba Cloud lists $1.50 per 1M input tokens and $4.50 output in Singapore, and $1.413 and $4.24 in China (Beijing), Hong Kong, Germany (Frankfurt), US (Virginia) and Japan (Tokyo).
  3. On Work360 at High effort, Alibaba says Qwen4-Max used 35% fewer tokens than Qwen3.8-Max and cut average cost per task by around 60%.

What happened

Alibaba has released Qwen4-Max, which it calls the most capable model in the Qwen family to date, for complex work, creative and research tasks. It is a mixture-of-experts model with 2.4 trillion total parameters that activates 71 billion per token, according to Alibaba's press release of 16 November 2026.

Alibaba's own benchmark claims are that Qwen4-Max ranked 6th on Code Arena, a real-world coding leaderboard (the release does not say against which models), and that on its Work360 evaluation it outperforms its predecessor, Qwen3.8-Max, overall. It says it did so with 35% fewer tokens at High effort reasoning mode and an average cost per task around 60% lower. These are Alibaba's results, not independent ones.

Qwen4-Max is available now on QwenCloud through the Token Plan and the API, and Alibaba says it can also be used across its platforms, including QwenWork and QwenStudio. Alibaba says open-weight, smaller models from the Qwen4 series will soon be released to developers worldwide, without giving a date.

It also says it filtered the training data by task intent, keeping defensive cybersecurity material and removing offensive attack content.

Alibaba Cloud's Model Studio documentation lists a context of up to 1,000,000 tokens, with a maximum output of 131,072 tokens without thinking and 262,144 for thinking plus the final answer.

The price is $1.50 per 1M input tokens and $4.50 per 1M output tokens in Singapore, with a free quota of 1 million tokens valid for 90 days. In China (Beijing), Hong Kong, Germany (Frankfurt), US (Virginia) and Japan (Tokyo) it is $1.413 input and $4.24 output.

An earlier Qwen4-Max-Preview, open only through a Token Plan from 11 November, will be retired after 25 November.

What it means for you

Our view

Read the $1.50 input price as Singapore's: the other five listed regions are priced lower, so what you pay depends on where you buy. The performance claims come from Alibaba's own Work360 evaluation and from a Code Arena ranking whose comparison set the release does not name, so they say little yet about your tasks.

Alibaba promises smaller open-weight Qwen4 models without a date. The earlier Qwen4-Max-Preview, open only through a Token Plan, will be retired after 25 November. Before moving any work, run a sample of your own tasks and compare cost per task in the region you would use.

It adds no new facts.

Share this story

Your reaction

We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.

Check our sources

Every sentence above is checked against these 3 sources.

1 Qwen4-Max-Preview, Model Studio documentationAlibaba Cloud · 8 facts Open the source archived copy
  1. Alibaba Cloud's documentation describes Qwen4-Max-Preview as the next-generation flagship multimodal model in the Qwen family, designed for coding, office productivity, frontend design and complex agentic tasks.

    Toned down to what the source says
    Qwen4-Max-Preview is the next-generation flagship multimodal model in the Qwen family, combining text generation, image and video understanding, and deep reasoning. It is designed for coding, office productivity, frontend design, and complex agentic tasks.
    cite
  2. Qwen4-Max-Preview takes text, images and video as input.

    Accepts text, images, and video for screenshot analysis, document understanding, and video analysis
    cite
  3. Qwen4-Max-Preview has a context length of 1M tokens and a maximum output of 393,216 tokens, shared by thinking and the final answer.

    Context length 1M tokens Maximum output 393,216 tokens, shared by thinking and the final answer
    cite
  4. Access to Qwen4-Max-Preview is through Token Plan Personal or Token Plan Team with a dedicated API key; the page states that pay-as-you-go access is not supported.

    Qwen4-Max-Preview is currently in preview and is available through Token Plan only. Pay-as-you-go access is not supported. ... Choose Token Plan Personal or Token Plan Team and follow its guide to obtain a dedicated API key.
    cite
  5. The preview period runs from 10:00 on 11 November 2026 to 24:00 on 25 November 2026 (UTC+8), and the model will be retired after the preview ends.

    The preview period runs from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8); model capabilities continue to be iterated and upgraded during the preview, and the model will be retired after the preview ends.
    cite
  6. The page says the production version, Qwen4-Max, is available and recommended for use.

    The production version, Qwen4-Max, is available and recommended for use.
    cite
  7. The page gives no price and no benchmark results for Qwen4-Max-Preview.

    Model ID: qwen4-max-preview
    cite
  8. The model is compatible with the OpenAI Chat Completions and Responses APIs and the Anthropic Messages API; thinking is enabled by default and reasoning effort can be set to low, medium or xhigh (default).

    API compatibility Compatible with OpenAI Chat Completions, Responses, and Anthropic Messages APIs ... Thinking: Enabled by default. Choose low, medium, or xhigh (default).
    cite
2 Qwen4-Max, Model Studio documentationAlibaba Cloud · 16 Nov 2026 · 3 facts Open the source archived copy
  1. Alibaba Cloud's Model Studio documentation describes Qwen4-Max (model ID qwen4-max) as the next-generation flagship multimodal model in the Qwen family, taking text, image and video input, with a context of up to 1,000,000 tokens; its page was last updated on 16 November 2026 and does not state a release date.

    Last Updated:Nov 16, 2026 Qwen4-Max is the next-generation flagship multimodal model in the Qwen family ... Model invocation ID (model parameter): qwen4-max ... Input modalities Text, image, video ... Context length 1,000,000
    cite
  2. Alibaba Cloud's Qwen4-Max page lists a price per 1M tokens of $1.50 input and $4.50 output in Singapore, with a free quota of 1 million tokens valid for 90 days, and $1.413 input and $4.24 output in China (Beijing), Hong Kong, Germany (Frankfurt), US (Virginia) and Japan (Tokyo).

    qwen4-max ... International Non-Thinking and Thinking modes 0<Token≤1M $1.5 $4.5 1 million tokens ... Valid for 90 days from the date of Model Studio activation, model release, or application approval, whichever is later ... China (Beijing) ... qwen4-max ... Non-Thinking and Thinking modes 0<Token≤1M $1.413 $4.24
    cite
  3. The Qwen4-Max page lists a maximum output of 131,072 tokens without thinking, and 262,144 tokens for thinking plus the final answer.

    Toned down to what the source says
    Maximum output length (non-thinking) 131,072 ... Maximum combined tokens for thinking and the final answer 262,144
    cite
3 Alibaba unveils Qwen4-MaxAlibaba Cloud · 16 Nov 2026 · 7 facts Open the source archived copy
  1. Alibaba's press release, datelined Hangzhou, 16 November 2026, says Alibaba has released Qwen4-Max, which it calls the most capable model in the Qwen family to date and says excels at complex tasks across work, creativity and research. the company says

    Hangzhou, China, November 16, 2026 - Alibaba has released Qwen4-Max, the most capable model in the Qwen family to date. ... Qwen4-Max excels at complex tasks across work, creativity, and research.
    cite
  2. Alibaba says Qwen4-Max is a mixture-of-experts model with 2.4 trillion total parameters that activates 71 billion per token, which it says is 25% fewer activated parameters than Qwen3.8-Max at the same 2.4T scale. the company says

    Built as an MoE model with 2.4 trillion total parameters, Qwen4-Max activates only 71 billion per token—25% fewer activated parameters compared to Qwen3.8-Max at the same 2.4T scale
    cite
  3. Alibaba says Qwen4-Max ranked 6th on Code Arena, a real-world coding leaderboard; the release does not say against which models. the company says

    It ranked 6th on Code Arena, a real-world coding leaderboard
    cite
  4. Alibaba says that on its Work360 evaluation Qwen4-Max outperforms its predecessor, Qwen3.8-Max, overall, using 35% fewer tokens at High effort reasoning mode and cutting average cost per task by around 60%. the company says

    According to Work360—a proprietary benchmark evaluating performance on challenging industry tasks across 721 specialized fields—Qwen4-Max outperforms its predecessor, Qwen3.8-Max, overall while using 35% fewer tokens at High effort reasoning mode and reducing the average cost per task by around 60%.
    cite
  5. Alibaba says Qwen4-Max is available now on QwenCloud via the Token Plan and API, and can also be used across Alibaba platforms including QwenWork and QwenStudio. the company says

    Qwen4-Max is available now on QwenCloud via the Token Plan and API. Users can also access the model across Alibaba platforms, including QwenWork and QwenStudio.
    cite
  6. Alibaba says open-weight, smaller-scale models from the Qwen4 series will soon be released to developers worldwide; the release gives no date. the company says

    Open-weight, smaller-scale models from the Qwen4 series will soon be released to developers worldwide.
    cite
  7. Alibaba says it filtered Qwen4-Max's training data by task intent, keeping defensive cybersecurity material and removing offensive attack content. the company says

    For cybersecurity, Alibaba filtered training data by task intent, retaining defensive material for vulnerability detection and remediation while removing offensive attack content.
    cite

Topics

The morning email

On the mornings we publish: the three top stories and up to four short ones. Free.

We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep