Alibaba opened Qwen4-Max-Preview on November 11, offering its next-generation flagship multimodal model for coding, frontend design and agent tasks. This is a limited preview, not general availability. The official documentation sets out its capabilities and access conditions.
The practical details
- Large working context: one million tokens, accepting text, images and video.
- Agent tools: tool calling, web search and structured JSON outputs.
- Access: model ID
qwen4-max-preview, available through Token Plan subscriptions only—not pay-as-you-go. - Time limit: the preview ends at the end of November 25, 2026 (UTC+8), after which this model will be retired.
Why it matters for builders
For SunMarc App Labs, a multi-file app fix or screenshot-to-frontend task would make a useful evaluation: measure correctness, tool reliability and review effort before changing the default model. We have not tested the preview.
Access restrictions matter. Token Plan Individual’s terms allow interactive use within coding and agent tools, but exclude automation scripts, custom application backends and non-interactive batch calls. Those Individual-plan conditions should not be confused with unrestricted production API access.
Existing clients also need an output-budget check: for this model, max_tokens covers reasoning and the final answer together, so an answer-only allowance may truncate responses. The compatibility notice explains the change; older Qwen models retain their previous behavior.
For another recent agent-model release and the importance of completed-task evaluation, see our Claude Fable 5.2 coverage.