Overview
- Snowflake announced dynamic model routing in Cortex AI Gateway as a private‑preview feature that automatically directs requests to the most cost‑effective, administrator‑approved model for each task.
- The company said routing decisions and model inference run inside Snowflake’s controlled perimeter, with each decision logged for audit and to enforce data residency and governance policies.
- Snowflake added DeepSeek‑V4‑Flash 0731 to private preview and announced GLM‑5.3 will be available in private preview soon, expanding the pool of open models the router can choose from.
- In internal tests reported by Snowflake, routing cut token use by up to three times on a dbt pipeline and by about 25% on a coding workload, though those results come from company benchmarks and private previews.
- The move reflects a wider industry shift toward using cheaper open models for well defined tasks, which Snowflake says could lower average AI cost per business outcome and make more workflows economical to automate.