Overview
decision-model-preview is a structured decision model that Alibaba Cloud added to Model Studio on 24 September 2026, listed in the QwenCloud model changelog. Alibaba describes it as built for high-frequency business decisions: it performs classification, yes-or-no decisions and scoring in parallel over text or business context, and returns probability distributions and confidence scores instead of free text. The use cases it names are ticket routing, content moderation, agent routing and result validation.
It is called through a separate Decision Model API rather than the chat endpoint: POST /compatible-mode/v1/systemone on a Model Studio workspace host in Singapore or China (Beijing). A request carries the model name, a state (the business context, as a string, object or array) and a map of questions. Each question has one of three types: choice (pick one option), noul (yes/no) or score (an ordinal scale). For each question the response gives the answer, the probabilities per option or level and, for choice and score questions, a confidence. Alibaba's API page also documents the TypeSafe SDK (typesafe-sdk) for calling it.
Model Studio lists a context window and maximum input of 65,536 tokens, text input, and a structured decision result as output. Function calling, structured outputs, web search, context caching, batch inference and fine-tuning are not supported. The API page recommends up to 16 questions per request and allows up to 255 options per choice question and 2 to 255 levels per score question. At launch the model was free for a limited time in both regions, with rate limits of 1,200 requests and 2,000,000 tokens per minute.
| Released | 2026-09-24 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Context | 65,536 tokens |
| Modalities | Text |
| Status | Preview |
Strengths
- Answers several typed questions about one input in a single call
- Returns probabilities and a confidence, not free text, so results can be thresholded in code
- Three question types: choice, yes/no (noul) and score
- 65,536-token context for long business records
- Free for a limited time at launch
Best for
- Routing support tickets to the right team
- Content moderation decisions
- Choosing which agent or tool should handle a request
- Checking whether an agent's result meets criteria
How to access
| Provider | Model ID |
|---|---|
| Alibaba Cloud Model Studio ↗ | decision-model-preview |
FAQ
What does decision-model-preview return?
For each question in the request it returns a typed answer: the chosen option for a choice question, a yes probability for a yes/no (noul) question, or a probability-weighted level for a score question. It also returns the probabilities per option or level and, for choice and score questions, a confidence.
How much does decision-model-preview cost?
At launch Alibaba Cloud Model Studio listed it as free for a limited time in both the Singapore and China (Beijing) regions.
How do I call it?
Send a POST request to /compatible-mode/v1/systemone on your Model Studio workspace host, with the model set to decision-model-preview, a state holding the business context and a map of questions. Alibaba's API page also shows the TypeSafe SDK, installed with pip install typesafe-sdk.
How much context can it take?
Up to 65,536 tokens of input. Its output is a structured decision result rather than generated text.