Snowflake has announced the introduction of dynamic model routing within Cortex AI Gateway and its flagship AI products.
The US cloud data platform is also expanding access to open source models in an effort to help enterprises reduce unnecessary AI expenditure and increase what it terms ‘intelligence efficiency’.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Dynamic model routing enables Cortex AI Gateway to automatically select the optimal AI model for each task, balancing quality and cost. This functionality extends across Snowflake’s products, including Snowflake CoCo, Snowflake CoWork, and third-party AI agents integrating with Cortex AI Gateway.
The system directs lower-complexity or repetitive tasks to more efficient models, reserving advanced frontier models for operations requiring deeper reasoning. Snowflake states this approach aims to help users reduce inference spending and minimise effort related to manual model selection.
Snowflake CEO Sridhar Ramaswamy said: “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.
“Achieving intelligence efficiency requires the flexibility to use the best model for each task as the landscape evolves. Snowflake’s role is to absorb that complexity so customers can focus on outcomes while we optimise model choice underneath.”
The company is also broadening access to open models, with DeepSeek-V4-Flash 0731 and GLM-5.3 set to join its portfolio of available models through Snowflake Cortex AI. These additions are designed to give customers more options to balance model performance and cost, while keeping governed data secure within the Snowflake platform.
The model library includes options from providers such as Anthropic, OpenAI, Google, SpaceXAI, Meta, and Mistral.
Dynamic model routing allows organisations to define which models and providers are accessible, a feature particularly important for those with specific regional or compliance requirements. As model performance and pricing change, routing decisions in Snowflake’s AI products like CoCo and CoWork can be updated automatically, allowing enterprises to adapt without having to rebuild their application infrastructure.
According to Snowflake’s internal testing, using a mix of open and proprietary models through dynamic model routing can maintain consistent output quality while materially increasing token efficiency.
In one case, agents using dynamic model routing with Cortex AI Gateway built a dbt pipeline with up to three times greater token efficiency than a frontier-model-only approach, delivering equivalent quality. In a separate test, engineering teams completed the same number of pull requests with 25 percent greater token efficiency.
Snowflake’s AI Research Team evaluated DeepSeek-V4-Flash on enterprise-focused tasks, with results indicating that DeepSeek v4 Flash scored 74.4% on data engineering tasks and outperformed proprietary models in the evaluation group.
GLM-5.2, also tested, scored 62.8% and used fewer tokens than any other model tested. The company believes these results indicate that open models are increasingly capable of supporting mission-critical enterprise workloads at lower cost.
In addition to dynamic routing and expanded model selection, Cortex AI Gateway provides administrators with greater visibility and control over AI resource use. Features include monitoring token usage and costs, setting spending limits, and allocating costs across various teams and AI agents.
Snowflake CoCo extends these controls, enabling administrators to set default models, attribute usage to teams or cost centres, establish per-user quotas, and receive notifications as consumption nears defined limits.
