Alibaba Cloud made its flagship model, Qwen4-Max, available through Model Studio on November 16, 2026. The new version comes with a one-million-token context window and lower list prices compared with the previous Qwen3.8-Max. According to the company’s Singapore price list, the model is priced at $1.50 per million input tokens and $4.50 per million output tokens. For Qwen3.8-Max, those figures were $2 and $6 respectively, which means a 25 percent discount on both counts.
Free quota and multimodal input
In the Singapore region, a free quota of one million tokens is also provided for nine days. The start of the quota is calculated based on whichever is latest among Model Studio activation, the model’s release date, or application approval.
According to the model documentation, Qwen4-Max takes text, image, and video input and generates text. Supported features include streaming, web search, parallel tool calling, function calling, prefix completion, and structured JSON output. Developers can access the model through OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Alibaba Cloud’s own DashScope interface. That makes it easier to integrate into existing coding and agent software.
The thinking budget comes out of the answer
Alibaba Cloud positions the model for coding, office productivity, interface design, and complex agent tasks. The thinking feature is enabled by default and runs at a high reasoning level; low, high, and very high (xhigh) options are also available. The key point to watch is this: the default thinking budget of 131,072 tokens is the same as the maximum output in non-thinking mode. On top of that, the thinking phase and the final answer share a single total output budget of 262,144 tokens. In other words, every token the model spends on reasoning reduces what is left for the answer.
Cache determines input cost
A price cut alone does not reduce the total bill. According to the Qwen4 cache guide, the same input prefixes, such as system prompts, tool definitions, and reference documents, can be reused for the same user. An explicit cache prefix must contain at least 1,024 tokens, and you can choose a lifetime of either five minutes or one hour. In Singapore, cache reads cost $0.075 per million tokens. Creating an explicit cache costs $1.875 for a five-minute lifetime and $3 for a one-hour lifetime. The discount applies only to input tokens; output tokens remain at the standard price.
For agents that repeatedly send long instructions and documents, prefix stability and cache lifetime therefore determine whether the discount actually shows up on the bill.
Regions and subscription limit
The model is listed in the Beijing, Singapore, Frankfurt, Virginia, Tokyo, and Hong Kong regions. Batch inference is supported only in Beijing among those regions, while fine-tuning and custom model deployment are not offered in any of the documented regions. In practice, control over the managed model therefore comes through API configuration, reasoning settings, and cache usage.
Qwen4-Max is also included in the Personal Edition Token Plan subscription in Singapore. The plan is billed with credits and supports interactive tools such as Claude Code, Cursor, Qwen Code, Qoder, and OpenClaw. However, the subscription is limited to interactive coding and agent tool usage only: automation scripts, application backends, and non-interactive batch calls are prohibited under the plan. That is an important difference that sets the subscription apart from pay-as-you-go API access.
The new version arrived after Qwen4-Max-Preview, which was made available only through the Token Plan on November 11, 2026. The preview version will remain active until November 25, 2026, after which it will be retired.
Source: Unite.AI

Share your experience
Bir sorunun, farklı bir çözümün veya eklemek istediğin bir notun var mı? Giriş yapmadan da yazabilirsin.