Alibaba Cloud opened access to Qwen4-Max-Preview on November 11, 2026, through its Token Plan service. The preview runs until the end of November 25, UTC+8, and is limited to interactive coding and agent tools. For university researchers, the release offers a temporary model to evaluate under access rules that exclude unattended batch workloads.
A preview with a defined retirement date
Alibaba’s Model releases and updates log lists Qwen4-Max-Preview under November 11. Its model documentation gives the precise window: November 11 at 10:00 through November 25 at 24:00, UTC+8. Alibaba says the model will remain unchanged during that period and will be retired afterward.
This is preview availability rather than general availability. The documented access conditions are:
| Condition | Documented position |
|---|---|
| Release status | Qwen4-Max-Preview, a temporary preview model. |
| Access channel | Token Plan only; pay-as-you-go access is unsupported. |
| Continuity | No model changes during the preview; retirement after it ends. |
| Workload scope | Interactive coding and agent tools; no automation scripts, custom application backends or non-interactive batch calls. |
The access and continuity conditions come from the model page; the workload limits are set out in Alibaba’s Token Plan permissions and usage policy.
Model capabilities do not expand the permitted workload
Alibaba documents text generation, image and video understanding, tool calling and structured outputs. It also lists compatibility with OpenAI Chat Completions, Responses and Anthropic Messages APIs. These are vendor-described capabilities, not an independent assessment of coding performance.
Token Plan’s usage policy permits calls initiated interactively through coding and agent tools. It prohibits automation scripts, custom application backends and non-interactive batch calls, and says misuse may lead to subscription suspension or an API key ban. Protocol compatibility therefore does not make an unattended research pipeline an allowed use of this plan.
For research software teams, the immediate opening is evaluation within an interactive development workflow, where generated changes can enter the team’s code review. An automated model evaluation suite that runs non-interactive batch calls falls outside the stated scope.
What the deadline means for universities
The retirement condition creates a practical limit for a reproducible AI experiment: later researchers cannot assume this hosted preview will remain available for reruns. Recording the model identifier, evaluation date, prompts, settings and outputs would preserve evidence of the configuration used, without guaranteeing future access to it.
Research offices also have an access and data-governance question to resolve. The Personal Edition overview restricts use to the subscriber and prohibits sharing. It describes Singapore as the available region but global model inference, with cross-border transfer of prompts and outputs. For an institutional pilot, those documented conditions matter alongside model capability and the short evaluation window.
Sources
- Alibaba Cloud Model Studio: Model releases and updates. Checked 2026-11-14.
- Alibaba Cloud Model Studio: Qwen4-Max-Preview model documentation. Checked 2026-11-14.
- Alibaba Cloud Model Studio: Token Plan permissions and usage policy. Checked 2026-11-14.
- Alibaba Cloud Model Studio: Token Plan Personal Edition overview. Checked 2026-11-14.
