Positioning and release
Released on September 24, 2026, Decision-Model-Preview is a preview structured decision model for high-frequency business decisions. Alibaba's fine-tuning guide identifies decision-model-preview-2026-09-24 as a model name; the general decision API documentation also uses the shorter name decision-model-preview. (docs.qwencloud.com)
Capabilities and uses
Given text or serialized business state, the model can evaluate multiple classification, yes/no and scoring questions in one forward pass. It returns structured decisions and probability distributions; confidence is also returned for choice and scoring questions. Intended uses include ticket routing, content moderation, agent routing and result verification. (help.aliyun.com)
Limits and access
- Maximum input and context: 65,536 tokens. The model card lists maximum generated-text output as 0; the API returns structured decisions rather than free-form generated text.
- The System One API accepts up to 16 questions per request, up to 255 choice options, and 2–255 scoring levels. The official announcement lists availability in Singapore and China North 2 (Beijing).
- Alibaba's model card lists function calling, web search, prefix completion, context caching and batch inference as unsupported. Its international card lists fine-tuning as unsupported, while the newer China-site guide documents fine-tuning of the dated model subject to access authorization and regional conditions; verify availability for the relevant account and region. (alibabacloud.com)
Pricing
Checked November 16, 2026. The following are platform-specific listings, not currency conversions:
- China-site Model Studio, CNY: The official model card lists inference as limited-time free in Beijing and Singapore, but does not specify separate input/output rates or a billing unit. The fine-tuning guide lists token-based training at CNY 0 per 1,000 tokens for a limited time; deployment and other training-resource charges may apply. (help.aliyun.com)
- International Model Studio, USD: The official international model card lists inference as limited-time free in Singapore and Beijing, without separate input/output rates or a billing unit. The QwenCloud model listing separately shows an input price of USD 0 per 1 million tokens; an output price is not listed. (alibabacloud.com)
Post-promotion inference prices and separate output billing are not verified. Check the applicable platform and region before use.