Qwen4-Max is generally available at $1.50 input, $4.50 output
The first Qwen4 model leaves the Token Plan preview with per-token prices 25% below Qwen3.8-Max.
Alibaba released Qwen4-Max on November 16, 2026, the first model of its Qwen4 generation, and opened pay-as-you-go API access as qwen4-max on Alibaba Cloud Model Studio. Until now the model was only reachable through the Token Plan preview that opened on November 11.
What changed
- New model ID
qwen4-max, billed per token in six regions. The preview IDqwen4-max-previewstays Token Plan only and is retired after 24:00 on November 25, 2026 (UTC+8). - Alibaba describes it as a sparse mixture-of-experts model with 2.4 trillion total parameters and 71 billion active per token, with native vision. Input is text, image or video; output is text.
- Context is 1,000,000 tokens. Output without thinking is capped at 131,072 tokens; thinking plus the final answer share 262,144 tokens (the preview card listed 393,216).
- It works with OpenAI Chat Completions, Responses, Anthropic Messages and DashScope, with function calling, web search, structured outputs and explicit and implicit caching. Batch inference and fine-tuning are not supported.
- Thinking is on by default at
reasoning_efforthigh; the other values arelowandxhigh. - No weights for Qwen4-Max. Alibaba says smaller open-weight Qwen4 models will follow.
Alibaba list prices
Per million tokens, from Model Studio's pricing page on November 16, 2026:
| Region | Input | Output |
|---|---|---|
| Singapore (International) | $1.50 | $4.50 |
| Beijing, Hong Kong, Frankfurt, Virginia, Tokyo (Global) | $1.413 | $4.24 |
In Singapore, cache reads are $0.075 and explicit cache creation is $1.875 (5-minute) or $3 (1-hour). Qwen3.8-Max lists at $2 input and $6 output in Singapore, so Qwen4-Max is 25% lower on both there. Singapore accounts also get a 1 million token free quota, valid for 90 days.
Qwen on RouterPlus
Qwen4-Max is not in the RouterPlus catalog as of November 16, 2026. Its predecessor is listed as qwen/qwen3.8-max. Per million tokens, from the catalog on November 16, 2026:
| Provider | Input | Cached input | Output |
|---|---|---|---|
| DeepInfra | $1.65 | $0.206 | $4.951 |
See the Qwen3.8-Max model page and all Qwen models on RouterPlus.
What to do
If you tested the preview on a Token Plan key, move to the GA ID before November 25. The GA model is billed per token on a pay-as-you-go key for your region.
model: qwen4-max-preview -> qwen4-max