Global – October 27, 2026 -- Snowflake has released snowflake-decision-fast in private preview, a lower-latency decision model whose endpoint round trips consistently complete in under 300 ms, according to the company. The release comes two weeks after Snowflake introduced the original Snowflake Decision in Cortex AI Functions.
Snowflake targets latency-sensitive workflows with three new capabilities
Customers asked for faster responses in live routing steps, real-time fraud checks and an agent's first triage hop, Snowflake said. The preview adds snowflake-decision-fast, REST API access through Cortex AI Gateway, and native availability of the new model through AI_COMPLETE in SQL. All three run under the existing governance, access-control and audit policies.
The models are built for classification, scoring, filtering and routing. They return typed answers, such as a category, a numerical score or a yes/no probability, rather than open-ended text.
Demo app makes three simultaneous decisions on every keystroke
Snowflake built a support-ticket application that evaluates the live text as a customer types. It routes the ticket to a team, flags urgency and scores sentiment at the same time. Options are constrained to valid choices, so the model cannot return an invalid action. That removes the need for parsers, retry loops or complex error handling.
The model returns per-option probabilities instead of free-text reasoning. Snowflake cites a 0.99 confidence on a "Billing" route as a signal workflows can threshold on.
SQL access extends the same logic to tables of millions of tickets
Through AI_COMPLETE, teams can apply the same decision logic across entire tables without building separate inference pipelines or managing external API integrations. Snowflake's example query classifies rows from a 10-million-row incoming support ticket table by routing queue and escalation need. It uses a global support platform triaging millions of tickets daily as the use case. Snowflake argues that a general-purpose LLM adds complexity, cost and processing overhead for narrowly defined classification.
Snowflake withholds benchmark figures for the fast variant
Snowflake compared snowflake-decision and snowflake-decision-fast on average accuracy and row-processing time using an internal decision benchmark, but the source gives no numerical results. The company said it expects ongoing model-quality updates in the coming months while working to cut latency and compute costs.
Access to the private preview requires a request through a Snowflake account team, which will follow up with documentation.