Alibaba opens Qwen4-Max to pay-as-you-go API users
Edited and reviewed by AI Ping Editorial Team
3-line summary
- Alibaba released the full Qwen4-Max model on its metered API.
- Singapore list rates are $1.50 per million input tokens and $4.50 per million output tokens.
- The subscription preview ends at 1 a.m. on November 26 in South Korea.
Alibaba released its flagship Qwen4-Max model on November 16. Developers can now call the full model through Alibaba Cloud Model Studio and pay for the tokens they use without a Token Plan subscription. In the Singapore International deployment, the standard rates are $1.50 per million uncached input tokens and $4.50 per million output tokens.
The November 11 preview used the model ID `qwen4-max-preview` and was available to Token Plan subscribers. The full release has its own ID, `qwen4-max`. Alibaba also added it to both the Personal and Team versions of Token Plan, giving subscribers and metered API customers access to the same full model.
A single request can include text, images and video, and the model responds in text. An application could send a product clip together with a catalog entry and ask the model to flag discrepancies in writing. Alibaba lists API formats compatible with OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. Billing and deployment are handled by the service receiving the request.
Alibaba lists a one-million-token context window. With thinking enabled, the Model Studio default, a request can take up to 983,616 input tokens and generate up to 262,144 tokens of reasoning and answer combined. With thinking disabled, the ceilings are 991,808 input tokens and 131,072 output tokens. Those separate ceilings determine how much material an application can send and how long a response it can request.
Singapore list rates are 25% lower than the previous model's
In the same Singapore deployment, Qwen3.8-Max costs $2 per million uncached input tokens and $6 per million output tokens. The full Qwen4-Max reduces each of those rates by 25%. A request's actual cost depends on token consumption. When the model runs in its default thinking mode, billed output includes reasoning tokens as well as the answer the user sees.
Repeated material that registers a cache hit qualifies for a separate read rate. In Singapore, that is $0.075 per million cached input tokens. Applications that resend the same brief or instructions can account for cached and uncached input separately. For a request with 100,000 uncached input tokens and 10,000 output tokens including reasoning, the standard Singapore token charge comes to $0.195.
Region also affects the price and available features. Alibaba lists a Singapore International deployment, a Beijing deployment for mainland China and Global deployment locations in Tokyo, Frankfurt, Virginia and Hong Kong. The Global deployment, which includes Tokyo, lists $1.413 per million input tokens and $4.24 per million output tokens. Alibaba's model card lists batch inference and fine-tuning as unavailable for the Singapore International deployment.
Vercel has added the model to its AI Gateway under the ID `alibaba/qwen4-max`. The gateway's November 15 changelog announced availability, and Vercel lists the model at $1.50 per million input tokens and $4.50 per million output tokens. The gateway and Alibaba's direct API each use their own model identifier.
In Alibaba's Work360 evaluation, the company said Qwen4-Max at high reasoning effort used 35% fewer tokens and lowered average cost per task by about 60% compared with Qwen3.8-Max. The figures describe the tasks Alibaba ran for that evaluation. The 25% decline in the Singapore price table comes from comparing what each model charges for the same number of input or output tokens.
Alibaba says `qwen4-max-preview` will close at the end of November 25 in China Standard Time, which is 1 a.m. November 26 in South Korea. Applications that still specify the preview ID have a cutoff for moving to the full release. The output ceiling for reasoning and answer combined also changes, from 393,216 tokens on the preview to 262,144 on the full model.
Sources
- Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery — Alibaba Cloud
- Qwen4-Max — Alibaba Cloud Model Studio
- Model pricing — Alibaba Cloud Model Studio
- Qwen4-Max-Preview — Alibaba Cloud Model Studio
- Token Plan product updates — Alibaba Cloud Model Studio
- Qwen4-Max now available on AI Gateway — Vercel
- Qwen4-Max on AI Gateway — Vercel
