Snowflake says AI model routing can cut costs while maintaining quality

Snowflake says businesses can reduce AI costs by routing each task to a model suited to its needs, rather than relying on the most powerful model for every request. The company describes this approach as “intelligence efficiency”: turning models, data and computing resources into measurable business outcomes.

Grace Adamson, AI Product Marketing Manager at Snowflake, told TechRadar that some tasks need advanced reasoning, while repetitive work such as classification, summarization and information extraction can be handled by more efficient models, including open-source options. Snowflake’s Cortex AI Gateway can route requests based on quality, speed and cost, while applying the same governance controls, including masking, to open and proprietary models. Customers can also limit which models and providers are available.

In internal testing cited by Snowflake, agents using dynamic model routing built a dbt pipeline with up to three times greater token efficiency than an approach using only a frontier model, while maintaining the same quality. The company says the savings from lower token use and inference costs can add up when processing millions of requests per month.

Snowflake argues that businesses should weigh model capability, speed and cost together. It says open models are improving for common enterprise tasks, while flexible model selection can help companies adapt as options change. These claims and test results are Snowflake’s; the source does not provide independent verification.