Qwen · Model guide
Qwen4 Max
The first model of the Qwen4 generation: a 2.4T-parameter mixture of experts activating 71B per token, plus a 328B n-gram embedding table, built on a hybrid Gated DeltaNet and Qwen Sparse Attention stack. It takes text, image and video in across a 1M-token context, with adjustable reasoning, function calling, structured output and context caching. Qwen aims it at professional work, coding and long-horizon agents that keep delivering results over many hours.
What this model is good at
What the vendor positions it for, and which of our models to reach for instead.
VendorQwen presents Qwen4-Max as an always-on AI teammate for work, creation and exploration, with stronger professional tool use, long-horizon task execution and delivery of finished results.source
Professional and industry work · Qwen reports its gains across finance research, legal due diligence, clinical records and engineering workflows, and a higher score than Qwen3.8-Max on its Work360 industry benchmark.(vendor claim)source
Long-horizon agents · Qwen positions the model for 24/7 agents that deliver a usable first version and keep shipping updates, and demonstrates game development in Unreal Engine, Unity and Godot.(vendor claim)source
Million-token context · Serves a 1M-token context with up to 991K input, 393K output and a 262K thinking budget.source
- qwen/qwen3.8-maxpredecessor —Choose Qwen3.8 Max to keep an existing integration on the previous generation's behavior.source
Pricing and billing
Billed per token. Cached input is charged at the cache-read rate.
| Price / 1M tokens | You pay |
|---|---|
| Input | $ 1.60 |
| Output | $ 4.80 |
| Cache read | $ 0.080 |
| Cache write | $ 2.00 |
+ 2.0K × $ 4.80
Capabilities and limits
What CrossModel guarantees across every route this model can take right now.
- Context window
- 1.0M tokens
- Max output
- 393.2K tokens
- Input / output modalities
- Text + Image + Video → Text
- Streaming
- Supported
- Tool calling
- Supported
- Structured output (JSON)
- Supported
- Image input
- Supported
- Thinking mode
- Supported· can be turned off
- Reasoning effort
- low · high · xhigh
- Thinking budget
- 1 – 262.1K tokens
- Available endpoints
- /v1/chat/completions · /v1/responses · /v1/messages
Published benchmarks
Scores the vendor reported, with the evaluation setup each one came from.
- DeepSWE 1.1resolved_rate·Qwen Team · 2026-11-16setup not fully disclosed
厂商自测;发布页图表未披露推理档位、harness 与采样设置。同图 GPT-6.1 Sol(75.2)、GPT-6 Astra(74.1)、Kimi K3.1 Pro(75.5)与各自厂商发布值一致。
75.7% - ProgramBenchscore·Qwen Team · 2026-11-16setup not fully disclosed
厂商自测(黑盒软件复刻);来源图表只给出数值,未说明具体 metric,保守记为 score;未披露评测设置。
31.5%
Vendor-reported numbers, not CrossModel measurements. Scores are only comparable when the benchmark version, metric and evaluation setup match, so nothing here is averaged or ranked.
Use it in your tools
Point the base URL at CrossModel and paste this model ID — every tool below has a setup guide.
Frequently asked questions
What is Qwen4 Max?
How much does Qwen4 Max cost?
Does Qwen4 Max support tool calling and structured output?
Which endpoint do I call?
Can I try it without writing code?
Chat first, integrate later
Chat first, then wire it in once you like the answers.