Alibaba opens Qwen4-Max-Preview with a one-million-token context window

Alibaba Cloud has opened Qwen4-Max-Preview in Model Studio, giving developers access to a temporary preview model whose documentation lists a one-million-token context window and multimodal capabilities.¹ The preview began on 11 November 2026 and is scheduled to be retired at midnight on 25 November, according to the official QwenCloud documentation summarised in the Corpus record.¹

What Qwen4-Max-Preview offers

Alibaba's Model Studio documentation lists qwen4-max-preview as a supported model in its platform.¹ The verified launch details describe support for reasoning, visual understanding and text generation, with access through Token Plan credits. Alibaba is also running a limited-time promotion in which usage can consume as little as 20% of the standard rate, alongside an additional night discount.¹

The one-million-token context window is the headline capability. In practical terms, a context window that large lets a model work across substantially more source material in one request than a conventional chatbot session. That could make the preview useful for long-document analysis, codebase work and multimodal research, although the preview label and short availability window mean developers should not treat it as a stable production contract.

The Corpus record also identifies reported support for web search, function calling, built-in tools, structured JSON and compatibility with OpenAI and Anthropic API formats.² Those detailed capability claims are attributed to secondary reporting rather than presented here as independent benchmark results. Alibaba has not published benchmark scores, a parameter count, model weights, a normal per-token price or a full Qwen4 release date in the verified material.¹²

A very short preview

The timing matters. The preview is scheduled to end on 25 November, less than two weeks after its opening, and the documentation says the model will then be retired.¹ That makes Qwen4-Max-Preview more useful as a live test of Alibaba's next model direction than as a dependable foundation for a long-lived application. Teams experimenting with it should keep a fallback model ready and avoid assuming that preview pricing or API behaviour will survive the retirement date.

The model is part of an increasingly crowded race to offer larger contexts and tool-aware reasoning. A million-token headline is impressive, but the missing benchmark data matters more than the number on its own: context length does not tell us how accurately a model retrieves, reasons over or cites information buried deep inside a long prompt.

Our opinion

Qwen4-Max-Preview is worth watching because Alibaba has paired a huge context claim with multimodal and tool-facing features rather than presenting a larger window in isolation. But this is a test drive, not a finished product announcement. The two-week retirement window, absent benchmarks and incomplete pricing picture make the sensible editorial conclusion straightforward: experiment if you already use Model Studio, but wait for a stable successor before building around it.