Temporary access to Qwen4-Max for coding and agentic tasks

In preview through November 25, Qwen4-Max advertises a million-token context window. API access requires Token Plan, with usage restrictions.

API access to Qwen4-Max is opening for a limited period. The official preview documentation schedules its retirement after midnight on November 25, 2026, UTC+8. Available under the model ID `qwen4-max-preview`, this version is set to remain unchanged throughout the trial. No launch timeline is specified for the full Qwen4 family.

The company positions the model for complex coding, document work, and agentic tasks, in which an AI carries out a sequence of actions using tools. Early access has been announced in QwenWork, Qoder, Qwen APP, and QwenStudio. For developers seeking API access, the preview is restricted to Token Plan, with no standalone pay-as-you-go option. A long context window for documents and code

The model documentation lists a one-million-token context window. It specifies text, image, and video inputs, with text output. Visual analysis can therefore accompany requests involving code, documents, or interfaces. The catalog also lists support for videos up to two hours long and 2 GB in size.

Documented features include tool calling, web search, and structured JSON output. These capabilities describe the available interactions without measuring success on a complete project. The official pages reviewed provide no benchmark scores to substantiate the claimed improvements in coding and autonomous task execution. They also specify neither Qwen4-Max’s parameter count nor a release of its model weights. A subscription with restrictions on API usage

The individual Token Plan subscription starts at $6 per month for the Lite tier, with a promotional price of $3 for eligible new users. It includes 20,000 monthly credits. This balance does not correspond to a fixed number of tokens: consumption varies with the model, reasoning, tool calls, and volume processed.

The individual plan’s terms restrict usage to interactive sessions in compatible coding or agent tools. They exclude automated scripts, noninteractive batch processing, and custom application backends. The subscription therefore does not provide unrestricted access for powering a service in the background.

The cloud region available for Token Plan is Singapore, with a dedicated API endpoint and keys. This setup does not restrict customers to residents of that country. However, the individual plan’s terms specify a “Global” deployment, with inference performed globally and cross-border transfers of requests and responses. Reasoning shares the response budget

The integration supports the OpenAI Chat Completions, Responses, and Anthropic Messages interfaces. Reasoning is enabled by default, with three available levels: `low`, `medium`, and `xhigh`, the last being the default setting. The reasoning documentation also describes the option to disable it for this model family.

The advertised maximum output is 393,216 tokens, but this limit is shared between reasoning and the final answer. With this preview, the Chat Completions `maxtokens` parameter now limits their combined total. An integration that previously reserved this budget for the answer alone must therefore leave room for reasoning to avoid receiving truncated text.