Qwen4-Max enters a time-limited preview through Alibaba’s Token Plan
Alibaba has opened Qwen4-Max-Preview, moving its next model generation beyond the release-watch stage. Official QwenCloud documentation puts the start at 10 a.m. Beijing time on November 11, or 02:00 UTC, and says the preview will end at the end of November 25 in Beijing (24:00 UTC+8). The model is scheduled to retire after that window. QwenCloud model guide
The hosted model accepts text, images and video and supports reasoning, tool calls and structured output. QwenCloud lists a one-million-token context window and a maximum output of 393,216 tokens, shared between reasoning and the final answer. Access through the developer platform requires Token Plan; pay-as-you-go calls are not supported during the preview. These are published specifications, not independent measurements of task performance. QwenCloud model guide
There is a practical compatibility change for developers switching models. Starting with this preview, the max_tokens parameter counts reasoning and answer tokens together. An allowance previously sized only for the visible answer can therefore cut a response short. QwenCloud says older Qwen models retain their existing behavior. API compatibility notice
The plan also has use restrictions worth checking before testing an agent. The individual subscription is intended for one person’s interactive development in supported tools, not production automation, batch scripts or scheduled background jobs. That makes the preview suitable for evaluation within the plan’s terms rather than an automatic replacement for a production API. Token Plan FAQ
This update confirms a named hosted preview. It does not establish that downloadable Qwen4 weights or the rest of the planned family have launched. No independent benchmark result is being asserted here.
Sources and citations
Revision history
Published .