Updated 5 min read
Qwen4-Max is out: price, benchmarks and how to try it.
Alibaba’s Qwen4-Max is now live on QwenCloud at $1.50/$4.50 per million tokens, with a 1M context. Specs, benchmarks, access and what people build with it.
By The Model Atlas · 9 sources · First published
In this post8 sections
Qwen4-Max is out. Five days after its preview, Alibaba’s Qwen team launched the full Qwen4-Max model on 16 November 2026, calling it “your new AI teammate for work, creativity, and discovery”. The API is live on QwenCloud at $1.50 per million input tokens and $4.50 per million output tokens. Here is what Qwen4-Max is, what changed since the preview, how it scores, what it costs and how to try it.
What is Qwen4-Max?
Qwen4-Max is the flagship of Alibaba’s Qwen4 family and the successor to Qwen3.8-Max. Alibaba Cloud’s docs describe it as “the next-generation flagship multimodal model in the Qwen family”, combining text generation, image and video understanding and deep reasoning, and designed for coding, office productivity, front-end design and complex agentic tasks.
The Qwen team says it is built on a new Mixture-of-Experts (MoE) architecture with native vision and a 1M-token context window. Alibaba has not published a parameter count.
What’s new in the Qwen4-Max launch
The preview that opened on 11 November was available through Token Plan only. The launch changes that. According to the Qwen team and Alibaba Cloud:
- A public API with pay-as-you-go pricing. The model ID is `qwen4-max`, and Alibaba lists it in its Model Studio release notes for 16 November 2026.
- Lower prices than Qwen3.8-Max. The Qwen team says input, output and cached-input prices are all lower than Qwen3.8-Max, with implicit caching down by 70%.
- Fewer tokens per task. Qwen says Qwen4-Max outperforms Qwen3.8-Max while using up to about 50% fewer tokens, and that average cost per task drops by more than 60% in its internal evaluation.
- Broader real-world work. Qwen says it handles tasks including 3D modelling, game development, manufacturing, finance, law, healthcare and frontier scientific discovery, with stronger long-horizon execution and better use of professional tools.
- Smaller open models to come. The announcement says open weights of smaller Qwen4 models will be released, without a date.
Qwen4-Max price
QwenCloud’s model page lists these prices per million tokens:
- Input: $1.50
- Output: $4.50
- Implicit cache input: $0.075
- Explicit cache creation: $1.875 (5-minute) or $3 (1-hour); explicit cache read: $0.075
The Qwen team’s announcement also links Token Plan, its subscription for coding and agent tools, alongside the API.
Qwen4-Max specs and features
Alibaba Cloud’s Model Studio docs list:
- Context: 1,000,000 tokens, with up to 983,616 input tokens in thinking mode and 262,144 tokens for thinking plus the final answer combined.
- Inputs: text, images and video, with text output.
- Thinking on by default, with `reasoning_effort` set to `low`, `high` (the default) or `xhigh`.
- Tools: function calling and parallel tool calls, web search, prefix completion and structured output (JSON Object and JSON Schema).
- APIs: OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and DashScope.
- Caching: explicit and implicit caching. Alibaba’s docs list batch inference and fine-tuning as not supported.
Qwen4-Max benchmarks
Alibaba’s own announcement gives efficiency claims rather than a benchmark table. Independent leaderboards published first results on launch day:
- Code Arena: WebDev. Arena says Qwen4-Max landed at #6 with a preliminary 1750 points, 74 points above Qwen3.8-Max (0902) at 1676.
- Agent Arena. Arena says it debuts at #14 overall, up from #39 for Qwen3.8-Max, and is the top Chinese model on that board.
- Artificial Analysis Intelligence Index. Artificial Analysis says Qwen4 Max at xhigh reasoning scores 43, nine points ahead of Qwen3.8 Max and three points behind Kimi K3.1 Pro at 46, at about $3.27 per Index task, roughly 11% less than Kimi K3.1 Pro.
- Where it trails. Artificial Analysis says it scores 48% on Terminal-Bench 4.0 against 67% for Kimi K3.1 Pro.
Leaderboard scores are early and can move as more votes and runs come in.
How to try Qwen4-Max
The Qwen team says you can use Qwen4-Max in Qwen Studio, QwenWork and Qoder, or through the API on QwenCloud. For the API:
- Create a QwenCloud (Alibaba Cloud Model Studio) API key.
- Point an OpenAI- or Anthropic-compatible client at the QwenCloud base URL for your region.
- Set the model to `qwen4-max`. Thinking is on by default; lower `reasoning_effort` for faster, cheaper answers.
Because thinking and the answer share one output budget, leave room in your token limit for reasoning, as with the preview.
What people are building with Qwen4-Max
Developers began testing Qwen4-Max during the preview, and five builds are on The Model Atlas so far:
- An adjustable pelican on a bicycle: an animated take on the classic pelican test, where every part can be tweaked.
- A restaurant website with a menu, food gallery and table reservations, made in Qoder.
- A campaign brief desk app that checks deadlines across Beijing and US time zones.
- A rainy-night convenience store in 3D, with an honest verdict from its creator that the 3D modelling is only middling.
- Lightning in a bottle, a visual-effects scene.
New builds land on the Qwen4-Max model page as creators share them.
Who is Qwen4-Max for?
Going by Alibaba’s description, Qwen4-Max is aimed at developers and teams doing agentic coding, front-end and design work, office tasks and long multi-step projects who want a flagship model at a lower price per task than Qwen3.8-Max. With a public pay-as-you-go API, it can now back production apps, not just interactive tool use as in the Token Plan-only preview.
When was Qwen4-Max released?
Alibaba’s Qwen team opened a Token Plan-only preview on 11 November 2026 and launched the full model, with a public API on QwenCloud, on 16 November 2026.
How much does Qwen4-Max cost?
QwenCloud lists $1.50 per million input tokens, $4.50 per million output tokens and $0.075 per million implicitly cached input tokens. Qwen says all three are lower than Qwen3.8-Max.
What is the Qwen4-Max model ID?
`qwen4-max`. The preview used `qwen4-max-preview`.
What is the Qwen4-Max context window?
1M tokens. Alibaba’s docs list 262,144 tokens for thinking and the final answer combined.
Is Qwen4-Max open source?
No. Qwen4-Max is available through QwenCloud and Qwen’s apps. The Qwen team says open weights of smaller Qwen4 models will be released later.
Is Qwen4-Max better than Qwen3.8-Max?
Qwen says it outperforms Qwen3.8-Max with up to about 50% fewer tokens. Independent boards agree it is a step up: Arena ranks it 74 points higher in Code Arena WebDev, and Artificial Analysis scores it nine points higher on its Intelligence Index.
Can Qwen4-Max understand images and video?
Yes. Alibaba’s docs list text, image and video input, with text output.
Sources · 9
- Qwen4-Max launch announcement (@Alibaba_Qwen on X, 16 November 2026) (opens in a new tab)
- Qwen4-Max model page and pricing (QwenCloud) (opens in a new tab)
- qwen4-max (Alibaba Cloud Model Studio docs) (opens in a new tab)
- Model releases and updates (Alibaba Cloud Model Studio docs) (opens in a new tab)
- Qwen4-Max blog (Qwen) (opens in a new tab)
- Qwen4-Max in Code Arena: WebDev (@arena on X, 16 November 2026) (opens in a new tab)
- Qwen4-Max in Agent Arena (@arena on X, 16 November 2026) (opens in a new tab)
- Qwen4 Max on the Artificial Analysis Intelligence Index (@ArtificialAnlys on X, 16 November 2026) (opens in a new tab)
- Qwen4-Max preview announcement (@Alibaba_Qwen on X, 11 November 2026) (opens in a new tab)