Snowflake introduces dynamic model routing to optimize AI costs

Snowflake has announced the introduction of dynamic model routing within its Cortex AI Gateway, Snowflake CoCo, and Snowflake CoWork. This feature is designed to optimize AI costs and performance by automatically selecting the most appropriate model for a specific task based on quality, speed, cost, and customer preferences.

Under this system, simple or repetitive tasks are directed toward more efficient models, while frontier models are reserved for complex reasoning requirements. Internal testing indicates significant efficiency gains; for example, Snowflake reported up to three times higher token efficiency for dbt pipeline creation and a 25% increase in token efficiency for engineering pull requests compared to using frontier models alone.

Additionally, Snowflake is expanding its library of open-source models. DeepSeek-V4-Flash 0731 is currently available in private preview, and GLM-5.3 will be added soon. The company is also introducing new tools to help enterprises track AI usage, allocate costs, and set spending limits and quotas across different teams and agents.