Alibaba Cloud has made Qwen4-Max, the new flagship model in its Qwen family of large language models, available to developers through its API. Specifications and pricing for the model appear on the company's Model Studio documentation and on the Qwen4-Max page of QwenCloud, Qwen's model platform, both dated November 16. The model is aimed mainly at coding, office productivity, frontend design and agentic tasks.

Text, Image and Video Input With a 1-Million-Token Context

Alibaba Cloud describes Qwen4-Max as the "next-generation flagship multimodal model in the Qwen family." Its model ID is qwen4-max.

The model accepts text, images and video as input and produces text output. The context length is 1,000,000 tokens, with maximum input of 991,808 tokens without thinking mode and 983,616 tokens with it. The reasoning_effort setting, which controls how much the model thinks, offers three levels—low, high and xhigh—with high as the default.

Key capabilities listed in the Model Studio documentation include:

  • Streaming output and thinking mode (including thinking summaries)
  • Function calling and parallel tool calling
  • Web search, structured output and prefix completion
  • Explicit and implicit context caching

The documentation also states that batch inference, fine-tuning and model deployment are not supported.

Callable Through OpenAI- and Anthropic-Compatible APIs

The model can be called in four ways: OpenAI's Chat Completions and Responses APIs, Anthropic's Messages API, and Alibaba Cloud's own DashScope. Apps already written for the OpenAI or Anthropic API formats can call it by swapping the endpoint and model name.

The QwenCloud model page lists built-in tools including code execution (code_interpreter), web search, web page fetching and extraction, and image-based search. Search and code execution can therefore be handled on the API side.

The model is offered in six regions: China (Beijing), Singapore (International), Germany (Frankfurt), the US (Virginia), Japan (Tokyo) and Hong Kong. Developers in Japan can choose the Tokyo region.

Pricing Starts at $1.5 per Million Input Tokens in Singapore, $1.413 in Tokyo and Elsewhere

According to Model Studio's pricing table, billing is pay-as-you-go per million tokens, with no difference between thinking and non-thinking modes.

Item (per 1M tokens)Singapore (International)Beijing, Tokyo, Hong Kong, Frankfurt, Virginia
Input$1.5$1.413
Output$4.5$4.24
Cache write (5 min)$1.875$1.767
Cache write (1 hour)$3$2.827
Cache read$0.075$0.071

Cache reads cost about 5% of the input price. For workloads that repeatedly feed the model long documents or codebases, whether caching is used makes a large difference in cost.

Official Pages Do Not Provide Performance Comparisons

Qwen is a series of large language models developed by Alibaba Group. In Model Studio's model list, Qwen4-Max sits at the top of both the text generation and visual understanding categories, and it is presented as suited to complex reasoning and coding.

However, the QwenCloud and Model Studio pages do not disclose the parameter count, training methods, or benchmark results against other companies' models. Nor do these pages mention any open-weight release.

Our Take

The combination of a 1-million-token context, image and video input, and compatibility with major API formats is clearly built for use cases such as agents and coding assistance, where the model reads large amounts of material and uses tools. With availability in the Tokyo region as well, conditions are favorable for developers considering switching from, or using it alongside, other companies' top-tier models.

On the other hand, the lack of batch inference and fine-tuning is a constraint for workloads that need to process large volumes cheaply or adapt the model to in-house data. Its real-world usability will have to be judged together with performance information that has not yet been officially disclosed.

Summary

Alibaba Cloud has begun offering Qwen4-Max, the new flagship of its Qwen series, via API. It accepts text, image and video input, has a 1-million-token context length, and is priced in Singapore at $1.5 per million input tokens and $4.5 per million output tokens. It can be called through OpenAI- and Anthropic-format APIs and is available in six regions, including Tokyo.

Sources: