Snowflake Decision: A Decision Model in Cortex AI Functions

Many enterprise AI tasks involve classification, scoring, or routing decisions that require fast, structured responses rather than complex reasoning. General-purpose LLMs can handle these workloads but often at higher cost and latency. Snowflake Decision is designed specifically for simple, high-volume tasks, returning typed answers like categories, scores, or yes/no probabilities that workflows can act on directly.

Snowflake Decision is now available in private preview through AI_COMPLETE within Cortex AI Functions, integrating with existing governance and access controls. The model evaluates multiple questions in a single call, such as assigning support tickets to teams, scoring customer frustration, and assessing urgency, reducing the need for separate model calls.

In evaluations using the Jev Decision Index 0.2.1 benchmarks, Snowflake Decision achieved the highest quality score of 57.63 among 72 models tested on 29 benchmarks. The benchmarks cover reasoning, language, retrieval, tool use, and creative judgment, with scores calculated using the index’s chance-adjusted methodology for fair comparison.

Cortex AI Functions allow model calls to be executed directly in SQL via AI_COMPLETE, with Snowflake Decision supporting three question types: Choice, Score, and Yes or No. The model returns structured answers, including per-option probabilities and confidence values, enabling workflows to process high-volume decisions efficiently.