Swapping a model name is not enough to move a coding tool to Qwen4-Max-Preview. Its output limit includes reasoning, so a configuration that previously reserved enough room for an answer can now cut that answer short. Check that integration detail before treating an incomplete response as a capability failure.
QwenCloud's model documentation says the preview runs from November 11 through November 25, Beijing time. The build stays unchanged during that window and is retired afterward. API access requires Token Plan; pay-as-you-go access is unavailable.
A dedicated key is only the beginning
The documented route is to choose an Individual or Team subscription, obtain its dedicated API key, and configure the dedicated endpoint for the protocol your client uses. Set the model identifier to qwen4-max-preview. Existing subscribers should follow that setup rather than reuse a conventional pay-as-you-go configuration without checking it.
For Chat Completions, max_tokens now has the same meaning as max_completion_tokens: both cap reasoning and the final answer together. QwenCloud explicitly warns that limits previously sized for answers alone can cause truncation. Our advice is to inspect the stopping reason and complete output before raising the budget or deciding the model cannot finish the task.
Thinking is on by default, with xhigh as the default reasoning effort; low and medium are also available. Do not set reasoning_effort and thinking_budget together. Chat Completions enables preserve_thinking by default. In that setting, multi-turn clients must return all historical reasoning in the assistant messages' reasoning_content field, not concatenate it into content. Check whether your chosen client actually handles that requirement.
The entry subscription is not a general-purpose API license
The Individual plan page lists Lite with 20,000 Credits per month at $6 per month, or a limited-time $3 monthly price. That offer is restricted to accounts that have never purchased an Individual subscription and ends on November 30, 2026 at 23:59:59 (UTC+8). Consumption varies with the model, tokens, thinking mode and tool calls. This is a Credits allowance, not unlimited usage. Unused allowance does not roll over, and the page says Individual subscriptions currently cannot be canceled.
The usage restriction matters just as much as the price. Individual is for interactive work inside coding and agent tools. Automation scripts, custom application backends and non-interactive batch calls are excluded. Possession of an API key does not authorize those workloads.
For routine questions or summaries, there is no need to switch immediately: the official selection guide still names Qwen3.7-Plus as its default. Token Plan offers only the Singapore region and Global deployment, with inference performed globally and prompts and outputs involving cross-border data transfer. Confirm that project material may be uploaded before testing; a project copy does not remove that data-transfer risk. We would treat this preview as a temporary test target instead. Validate edits on a project copy, keep the previous model configuration, and arrange a switch back before access ends.