Alibaba opens Qwen4-Max-Preview on Token Plan through 25 November

The page says the model accepts text, images and video, including screenshot analysis, document understanding and video analysis, and that it generates text[1]. Alibaba Cloud's Model Studio page describes the same inputs[3]. AI.info reports that the model returns text[5].

Context length is 1 million tokens. Maximum output is 393,216 tokens, shared by thinking and the final answer[1][3]. Thinking is enabled by default. The page says developers choose reasoning_effort of low, medium or xhigh, with xhigh as the default, and that reasoning_effort and thinking_budget must not be set together[1].

It lists function calling, web search and built-in tools, plus structured outputs as JSON Object or JSON Schema. It says the API is compatible with OpenAI Chat Completions, OpenAI Responses and Anthropic Messages[1][3]. From this model, QwenCloud says max_tokens and max_completion_tokens both cap the combined tokens for thinking and the answer, so a limit sized for the answer alone can cut the reply short. It says earlier Qwen models keep their previous behaviour[1].

QwenCloud says the model is designed for coding, office productivity, frontend design and complex agentic tasks[1]. That is the description on the page.

The quick start on both the QwenCloud page and the Alibaba Cloud page says to take a dedicated API key from Token Plan Individual or Token Plan Team, and to set the model ID to qwen4-max-preview[1][3]. The base URLs on those pages differ. QwenCloud gives https://token-plan.maas.qwencloudapi.com[1]. Alibaba Cloud's Chinese Model Studio page gives https://token-plan.ap-southeast-1.maas.aliyuncs.com[3].

The Token Plan Individual page says the alias of qwen4-max-preview is qwen4.0-max-preview, that both IDs work, and that bills show qwen4.0-max-preview[6]. AI.info reports the same alias[5].

The Individual page lists Lite at US$6 a month for 20,000 Credits, with a limited-time new-user price of US$3. Essential is US$10 a month, or US$5 for new users, for 45,000 Credits. Standard is US$18 a month for 80,000 Credits. Pro is US$43 a month for 200,000 Credits. A Credit Pack is US$15 for 20,000 Credits, up to five at a time, and it requires an active subscription[6]. The page calls the Lite and Essential new-user prices limited-time, and it gives no end date for them.

For this preview, that page says Credits use during the promotion is as low as 20% of the standard rate, which it calls 5x usage. Calls from 22:00 to 08:00 (UTC+8) get an additional 60% off, which the page says is 8% of the standard rate[6]. Those hours are 01:00 to 11:00 AEDT. AI.info reports the same 20% and 8% figures[5]. Neural Brew reports 4% for that overnight window[4]. The Individual page says the Credits for one request depend on the model, token use, thinking mode and tool calls, and it points to the Token Plan usage view for the actual charge. It does not print a standard Credits price for one Qwen4-Max-Preview call, and it says QwenCloud may change the promotion[6].

The Individual terms limit use to interactive sessions inside programming and agent tools. The examples named on the page are Claude Code, Cursor, Qwen Code, Qoder and OpenClaw. The terms exclude automation scripts, custom application backends and non-interactive batch calls. They also say the only Token Plan region is Singapore, with Global deployment, so prompts and model outputs cross borders[6]. When the monthly quota is used up and no Credit Pack Credits remain, the page says the service pauses until the next subscription month. Unused quota does not carry into the next cycle[6].

The QwenCloud model page, the 11 November changelog entry and the Alibaba Cloud page give no benchmark scores and no parameter count, and they do not announce an open-weights download[1][2][3]. Neural Brew reports that the announcement includes no model card and no licensing details, and that open-weights availability, a pay-as-you-go price and any further Qwen4 sizes have not been announced[4]. The model page gives no date for a general release. The date it does publish is the retirement of this preview[1].

QwenCloud says this build stays unchanged until 24:00 on 25 November 2026 (UTC+8), 03:00 on 26 November AEDT, and is then retired[1]. Calling it requires a Token Plan. Individual plans on that page start at US$6 a month, with a new-user price of US$3 on Lite, and the page describes a temporary cut in Credits use for the preview[6]. QwenCloud calls the model its next-generation flagship. The model page leaves out a benchmark score, a parameter count and an open-weights release[1][4].