Snowflake Decision
⚠️ Why not a model
Private-preview Cortex model; base model, weights and training are undisclosed. Re-check if Snowflake documents a decision-trained model or opens it up.
Technical specs
- Base LLMs
- undisclosed
- Decision types
- choice, score, boolean
- Features
- hosted-api, private-preview, sql-ai-complete, jev-adapter, throughput-oriented
- License
- Proprietary (Snowflake Cortex, private preview)
Links
Snowflake Decision (docs, blog, release note 7 Oct 2026) is called from SQL as AI_COMPLETE(model => 'snowflake-decision', prompt => TO_JSON({state, questions})), in the Jev request shape, with choice, score and noul (true/false) questions. During the private preview an account needs CORTEX_ENABLED_CROSS_REGION = 'ANY_REGION', ENABLE_AI_COMPLETE_JEV_ADAPTER and AI_COMPLETE_JEV_ADAPTER_MODELS_LIST = 'snowflake-decision'. Access goes through the account team; it is not supported for production. Snowflake says it is built for throughput, and a single call can take seconds.
Numbers (UnverifiedClaim): 57.63 and “first among 72 models” on a 29-benchmark subset of Decision Index 0.2.1 (88,837 requests); Snowflake says interface limits stopped it from running the full suite. Not on the official 0.3 board.
Why a runtime entry. Snowflake does not say what model sits behind the endpoint, so there is nothing to score. Same treatment as OpenAI’s Decisions API.
