qwen3.8-max is Alibaba’s current production flagship. It replaced the preview route in August 2026 and is a substantial change from the older Qwen3 Max covered by this URL. It accepts text, images, and video, supports function calling and web search, and has a 1 million-token context window.
Current API prices
Alibaba prices the model by deployment region. These are the list prices shown in the model documentation on August 24, 2026:
Qwen3.8-Max price per 1M tokens
| Singapore | $2 input; $6 output |
|---|---|
| Germany, US, Japan, Hong Kong | $1.65 input; $4.951 output |
| Implicit cache hit in Singapore | $0.25 input |
| Implicit cache hit in listed global regions | $0.206 input |
| Explicit cache read in Singapore | $0.17 input |
| Explicit cache read in listed global regions | $0.137 input |
Promotions can change the invoice, so check the Model Studio console before making a budget. The table above uses published list prices, not a temporary sale.
Cache repeated context
Cache pricing matters when an agent repeatedly sends the same repository instructions or long document. In Singapore, an implicit cache hit costs one eighth of normal input. Explicit cache creation costs more than ordinary input, so it pays off only when the cached material is reused.
When to use a cheaper Qwen
Qwen3.8-Max is appropriate when a task needs its multimodal input, very long context, or stronger agent behavior. Routine text transformations and small code edits may cost less on a Plus, Flash, or smaller open model. Compare the exact model IDs and regional prices in Model Studio because availability differs by workspace location.
The Qwen team also released open Qwen3.8 weights. The official repository lists a 2.4T-A95B model and a 27B model. The 27B release is the more realistic local option, although quantization, context length, and hardware still determine whether it will run well.
Use the production ID
Set the model to:
qwen3.8-maxDo not start a new configuration on qwen3.8-max-preview. Alibaba’s August upgrade notice says preview traffic is routed to the official release.
For a coding workflow, see Qwen with Claude Code and Qwen Code CLI on Windows.
