Snowflake launches dynamic model routing to cut AI costs

Snowflake frames these announcements around what it calls intelligence efficiency: matching each task with a model that delivers required quality at the lowest appropriate cost, rather than defaulting to the most expensive option. The company argues that a router is only as good as the pool of models it can choose from, which is why it is expanding open model access alongside the routing capability.

Early internal testing supports the efficiency claim. Dynamic routing completed a dbt pipeline workload with up to three times greater token efficiency than a frontier-model-only approach, and a coding workload used about 25% fewer tokens. On ADE-bench, DeepSeek-V4-Flash scored 74.4%, outperforming the leading proprietary model Snowflake tested.

Snowflake also emphasized that it serves these open models itself rather than proxying a third-party API, so inference runs within a secure perimeter near governed data. This is intended to reduce transfer cost and latency while keeping data within Snowflake's governance boundary. The company positions these capabilities as compounding: as routing decisions adapt to new models and real-world usage, a greater share of workloads can be handled by efficient models over time.