S3DAT NEWS
· 8 stories RSS s3dat.com

Alibaba opens Qwen4-Max preview with a November 25 cutoff

Alibaba opened Qwen4-Max in preview on November 11 through Token Plan. The release gives developers a new model to evaluate inside coding and agent tools, with a scheduled end to this preview version.

Spec sheet

Model
qwen4-max-preview
Preview ends
25 November 2026 at 24:00 UTC+8
Access
Token Plan; no pay-as-you-go access
Personal plan region
Singapore; Global deployment
Permitted use
Interactive coding and agent tools

A fixed evaluation window

Alibaba’s model documentation describes text generation, image and video understanding, reasoning, tool calls and structured outputs. It targets coding, office work and tasks involving agents. The preview began at 10:00 on November 11, China time, and ends at midnight after November 25. That cutoff is 16:00 UTC on November 25. Alibaba says the model will remain unchanged during that period and retire afterward. That makes the preview an evaluation target with an expiry date. Results gathered against this version describe that version; the announcement does not turn it into a permanent production dependency.

Access comes through a tool subscription

The Personal Edition terms limit Token Plan to interactive use in compatible coding and agent tools. Automated scripts, application backends and non-interactive batches fall outside that scope. Alibaba lists reasoning, visual understanding and text generation for Qwen4-Max in its supported-model table. The same table retains Qwen3.8-Max and Qwen3.8-Flash. The preview therefore adds another model choice within the subscription instead of replacing the existing entries. The international Personal Edition operates from Singapore with Global deployment, and its terms disclose cross-border handling of prompts and outputs.

The connection is separate from standard billing

The November 4 quick-start guide describes a dedicated Token Plan key and two protocol options: OpenAI-compatible and Anthropic-compatible. Token Plan credentials cannot be exchanged with ordinary Model Studio or Coding Plan credentials. Each key must match its own endpoint. This separation matters when a developer already has a working client configuration: the model name alone does not determine the billing route. The guide lists clients including Cursor, Codex, Claude Code, Qwen Code and Qoder. For tools using the Responses API, its Harness integration can add search or code execution, with Credits charged for successful tool invocations.

A production endpoint is a different decision

Alibaba’s October 11 endpoint guide separates subscription traffic from pay-as-you-go traffic. It documents legacy shared endpoints, workspace-specific endpoints and trial endpoints for the latter. Alibaba recommends workspace-specific domains for production and describes lower rate limits for trial domains. Those routes have their own regional keys and workspace requirements. Access to the new preview does not confer access through every documented route: the model page explicitly excludes pay-as-you-go. For an existing application, this divides two decisions that often look similar in a model picker. One concerns evaluating a model in an interactive development tool. The other concerns serving repeated requests from a backend under a different access arrangement.

Claim check

  • Confirmed Alibaba documents a fixed Qwen4-Max preview ending on November 25, followed by retirement of that preview model. [1]
  • Confirmed Token Plan Personal Edition excludes automated scripts, custom application backends and non-interactive batch calls. [2]
  • Company claim The model’s coding, reasoning and multimodal abilities are described by Alibaba; these documents do not independently establish comparative performance. [1]
  • Unclear The timing, access terms and capabilities of a replacement after the preview are not established by the cited release documents. [1]

Sources