Snowflake adds dynamic model routing to Cortex AI Gateway

Snowflake is introducing dynamic model routing to its Cortex AI Gateway to help enterprises optimize AI costs and performance. The new feature will automatically select the most appropriate model for a given task based on quality, speed, and cost requirements.

Under this system, routine or low-complexity tasks will be routed to more efficient, lower-cost models, while complex reasoning tasks will be directed to frontier models. This approach aims to improve what Snowflake calls ‘Intelligence Efficiency,’ ensuring that compute and data resources are converted into business value effectively.

As part of this expansion, Snowflake is adding new open models to its library, including DeepSeek-V4-Flash 0731 and GLM-5.3. The dynamic model routing feature will be integrated across Snowflake CoCo and Snowflake CoWork and will also be available to third-party AI agents via the Cortex AI Gateway. The company noted that internal testing suggests using a mix of open and proprietary models can improve efficiency by up to three times.