What Is Qwen4-Max? Alibaba's Flagship: Price, Specs, Access
The short answer
Qwen4-Max is Alibaba’s new closed-weights flagship, launched November 16, 2026, priced at $1.50 input and $4.50 output per million tokens with a 1M-token context. It reads text, images and video, writes text, reasons by default, and is reachable through OpenAI- and Anthropic-compatible APIs as qwen4-max. It is the first model of the Qwen4 generation; Alibaba says it beats Qwen3.8-Max on its own Work360 benchmark (52.69 vs 47.74).
Specs at a glance
| Qwen4-Max (QwenCloud model page, Nov 17, 2026) | |
|---|---|
| Released | November 16, 2026 |
| Model ID | qwen4-max |
| Input / output | Text, image, video → text |
| Context | 1,000,000 tokens (max input 991K; 983K in thinking mode) |
| Max output | 131,072 tokens (262,144 in thinking mode) |
| Reasoning | On by default; reasoning_effort low / high / xhigh, default high, thinking budget 131,072 |
| Price | $1.50 input · $4.50 output · $0.075 cached input per 1M tokens |
| Explicit cache creation | $1.875 (5 minutes) · $3 (1 hour) per 1M tokens |
| Rate limit | 5,000,000 tokens per minute (default tier) |
| APIs | OpenAI Chat Completions and Responses, Anthropic Messages, DashScope |
| Built-in tools | Web search, web extractor/fetch, code interpreter, image search |
| Open weights | No |
What Alibaba says is new
- Work360, a benchmark Alibaba built on a taxonomy of professional work, scores Qwen4-Max at 52.69 against 47.74 for Qwen3.8-Max. No independent run of it exists yet, so treat it as a vendor claim.
- Alibaba also says Qwen4-Max needs fewer tokens per task than Qwen3.8-Max and is cheaper per task; per-task cost depends on how much your workload makes it think, so measure on your own prompts.
- The model targets coding, office work, front-end design and agent tasks, with tool calling, structured outputs, prefix completion and context caching.
Where it sits on price
Using our current API prices, per million input / output tokens:
| Model | Input | Output |
|---|---|---|
| GPT-6.1 Terra | $1.00 | $5.00 |
| GLM-5.3 (Z.ai) | $1.40 | $4.40 |
| Qwen4-Max | $1.50 | $4.50 |
| Gemini 4 Hydrogen | $1.50 | $7.50 |
| Kimi K3.1 Pro | $2.00 | $10.00 |
| Claude Sonnet 5.5 | $2.00 | $10.00 |
Qwen4-Max is among the cheapest flagship-class models on output, roughly half the output price of Kimi K3.1 Pro and Claude Sonnet 5.5. Because reasoning is on by default with a 131K thinking budget, output tokens — and the bill — can still run high on hard tasks; set reasoning_effort to low for routine calls.
How to get access
- QwenCloud (international): create an API key, call
qwen4-maxwith the OpenAI SDK or the Anthropic SDK by changing the base URL. Model Studio lists a 1-million-token free quota valid for 90 days for new users. - Alibaba Cloud Model Studio (Chinese mainland deployment): same model ID; its page marks batch inference and fine-tuning as not supported, while QwenCloud’s page lists batch and fine-tuning features — check the console for your region before planning either.
- Qwen apps and Alibaba’s Token Plan subscriptions for chat and coding tools rather than API use.
Who should try it
- Teams already on Qwen3.8-Max: the upgrade is a model-ID change at a lower list price.
- Cost-sensitive agent workloads that need vision and a 1M context: test it against GPT-6.1 Terra and Gemini 4 Hydrogen.
- Anyone who needs to self-host: look elsewhere — see how to choose an open-weight LLM to self-host.
Last verified: November 17, 2026.