Snowflake is adding dynamic model routing to Cortex AI Gateway. The feature automatically selects the model that best balances quality and cost. In addition, open-source models such as DeepSeek-V4-Flash 0731 and GLM-5.3 will become available within Cortex AI.
The routing feature directs simple or repetitive tasks to more efficient models, while routing work requiring deeper reasoning to frontier models. This means users no longer have to make that choice themselves for each request. The feature works in Snowflake CoCo, in CoWork, and in third-party AI agents that run through the gateway.
The Cortex AI Gateway itself is not new. Snowflake announced this layer in July 2026 as the foundation for managing agent connections, routing requests, and optimizing AI consumption. The company now refers to this as “intelligence efficiency,” that is, how effectively compute, models, data, and context translate into business results.
Three times the token efficiency
According to Snowflake, internal tests show that a mix of open and proprietary models delivers comparable quality with better token efficiency. In one evaluation, agents using dynamic model routing built a dbt pipeline with up to three times greater token efficiency than a route that used only Frontier models. In another test, engineering teams completed the same number of pull requests with 25 percent greater token efficiency.
Administrators, however, remain in control. They determine which models and providers are available to users, a factor that, according to Snowflake, is important for organizations dealing with regional model availability or compliance requirements. If a model’s price or performance changes, the gateway adjusts routing decisions without requiring apps or agents to be rebuilt.
DeepSeek-V4-Flash 0731 and GLM-5.3 will be available via Cortex AI, CoCo, and CoWork. They join models from Anthropic, OpenAI, Google, SpaceXAI, Meta, Mistral, and others.
“Enterprises are becoming much more rigorous about the economics of AI. The question is no longer how much AI they are using, but whether that AI is translating into meaningful business value,” says Snowflake CEO Sridhar Ramaswamy. According to him, it’s up to Snowflake to eliminate the complexity of model selection.