Oct 7, 2026: Snowflake Decision with AI_
With this release, we introduce the private preview of Snowflake Decision, a model available through
AI_COMPLETE that evaluates text and structured data against user-defined
questions. The snowflake-decision model supports classification, rubric-based scoring, and true-or-false assessment,
returning structured answers with probabilities for downstream application logic.
Snowflake Decision supports the following question types:
- Choice: Selects an option from a fixed set and returns the selected key, a probability for each option, and confidence.
- Score: Evaluates content against an ordered rubric and returns a probability-weighted score, a probability for each level, and confidence.
- True-or-false assessment: Returns the probability that a specified statement is true.
You can combine multiple assessments of the same content in one request and evaluate rows across a table or query result in a single SQL statement. Applications can use the returned probabilities and confidence values to route results to automated actions, human review, or a fallback.
For example, a support-ticket workflow can select the responsible department, score customer frustration, and assess urgency in one call.
For more information, see Snowflake Decision with AI_COMPLETE.