Qwen4-Max
Alibaba's API-only Qwen4 flagship with image and video input and a 1M-token context
Context window
1M
Input / 1M tokens
$1.50
Output / 1M tokens
$4.50
Provider
Alibaba
QwenCloud (Alibaba Cloud Model Studio) international (Singapore) list price, the same in thinking and non-thinking mode across the full 1M context; implicit cache hits cost $0.075 per 1M input tokens. Deployments in Beijing, Hong Kong, Germany, the US and Japan list $1.413/$4.24. The Singapore region includes a free quota of 1 million tokens, valid for 90 days. Batch inference and fine-tuning are not supported. · Data verified 2026-11-17
Qwen4-Max (November 16, 2026) is Alibaba's next-generation flagship and, per Alibaba, the most capable Qwen model to date. It has 2.4 trillion total parameters with 71B active per token, accepts text, image and video input with a 1M-token context, and thinks by default with low, high and xhigh reasoning effort. Unlike Qwen3.8-Max, whose text core was released as open weights, Qwen4-Max is API-only; Alibaba says smaller open-weight Qwen4 models will follow. Alibaba reports it beats Qwen3.8-Max on its Work360 benchmark while using 35% fewer tokens at High effort and cutting average cost per task by about 60%, and that it ranks 6th on Code Arena. According to The Next Web, Alibaba's launch charts put it at 75.7 on DeepSWE v1.1 and 48.5 on Terminal-Bench 4.0; TNW notes these are Alibaba's own numbers.
How to access it
QwenCloud / Alibaba Cloud Model Studio API (model qwen4-max) with OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and DashScope interfaces, plus the Token Plan subscription, Qwen Studio, QwenWork and Qoder. A Token Plan-only preview (qwen4-max-preview) opened November 11 and is retired after November 25, 2026. Weights are not released.
Strengths
- ✓Lower price than Qwen3.8-Max: $1.50/$4.50 vs $2/$6 per 1M
- ✓Text, image and video input with a 1M-token context
- ✓Up to 262,144 output tokens in thinking mode
- ✓OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and DashScope-compatible APIs
- ✓Alibaba-reported 35% fewer tokens than Qwen3.8-Max at High effort on Work360
Best for developers who...
When to choose it (and when not to)
Reach for Qwen4-Max when...
- →You want Alibaba's strongest model at $1.50/$4.50 per 1M, below Qwen3.8-Max's $2/$6
- →Multimodal agent work over images and video with a 1M-token context
- →You use OpenAI- or Anthropic-compatible tooling and want to switch models without rewrites
Look elsewhere if...
- ✕You need open weights or self-hosting (Qwen4-Max is API-only; Qwen3.8-Max's text core is open)
- ✕Batch inference or fine-tuning (not supported)
- ✕You need SWE-bench Pro, SWE-bench Verified or GPQA Diamond results: Alibaba has not published them for this model
How to use it
- ›Thinking is on by default; set enable_thinking to false for quick replies
- ›reasoning_effort accepts low, high (default) and xhigh; minimal maps to low, medium to high and max to xhigh
Quickstart
Pythonimport os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["DASHSCOPE_API_KEY"],
base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
)
resp = client.chat.completions.create(
model="qwen4-max",
messages=[{"role": "user", "content": "Plan the refactor of this module step by step."}],
)
print(resp.choices[0].message.content)International (Singapore) OpenAI-compatible endpoint, same pattern as the other Model Studio models.
API model id: qwen4-max
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Code Arena | Ranked 6th | Alibaba-reported at launch |
| Work360 | Beats Qwen3.8-Max with 35% fewer tokens (High effort) | Alibaba internal benchmark across 721 specialized fields; about 60% lower average cost per task |
Compare Qwen4-Max
Compare Qwen4-Max with any other model
Build a comparison →All model comparisons →