Snowflake Adds Dynamic Model Routing to Cortex AI Gateway

Snowflake announced in a press release dynamic model routing within Cortex AI Gateway to help enterprises select AI models based on cost and quality. The feature is designed to reduce unnecessary AI inference spend by matching each request with a suitable model.

Dynamic model routing is integrated across Snowflake CoCo and Snowflake CoWork, and it is available to external AI agents using Cortex AI Gateway. The system sends lower complexity or repetitive tasks to more efficient models and routes tasks that need deeper reasoning to frontier models.

Snowflake also plans to expand access to open models through Snowflake Cortex AI, including DeepSeek-V4-Flash 0731 and GLM-5.3. The model portfolio also includes providers such as Anthropic, OpenAI, Google, SpaceXAI, Meta, and Mistral.

In internal tests, agents using dynamic model routing with Cortex AI Gateway built a dbt pipeline with up to three times greater token efficiency than a path using only frontier models, while keeping the same quality. In a separate test, engineering teams completed the same number of pull requests with 25 percent greater token efficiency.