Alibaba released Qwen4-Max, the first Qwen4 model, after a five-day Token Plan preview

Alibaba Cloud’s Model Studio changelog lists qwen4-max as released on 2026-11-16, describing Qwen4-Max as “the next-generation flagship multimodal model in the Qwen family,” combining text generation, image and video understanding and deep reasoning, and aimed at coding, office productivity, frontend design and complex agentic tasks. It is the first model of the Qwen4 generation on the platform, and it is compatible with the OpenAI and Anthropic APIs.

The general release followed a preview. On 2026-11-11 Alibaba opened qwen4-max-preview through its Token Plan subscription only, with no pay-as-you-go access, for a window running from November 11 to November 25, 2026 (UTC+8). The preview page now notes that the production version, Qwen4-Max, is available and recommended.

The model documentation lists a 1,000,000-token context length, up to 131,072 output tokens in non-thinking mode and 262,144 tokens combined for thinking and the final answer. It takes text, image and video input and returns text, with web search, parallel tool calling and structured output. International pricing in Singapore is 1.5 dollars per million input tokens and 4.5 dollars per million output tokens for requests up to 1M tokens.

Why it matters: a Chinese lab’s new-generation flagship arrives with a 1M context priced at a fraction of US frontier rates - its output price is roughly a ninth of Claude Fable 5.2’s 40 dollars.

What it does not show: Alibaba’s documentation gives no benchmark table, parameter count, license or downloadable weights for Qwen4-Max, so there is no official evidence yet of how it compares with other frontier models or whether any Qwen4 weights will be opened.