Quick Facts
- Snowflake’s dynamic model routing delivered up to 3x greater token efficiency in benchmark tests while maintaining output quality.
- 73% of organizations reported AI costs exceeded original projections, according to the FinOps Foundation’s State of FinOps 2026 report.
- Snowflake added two new models to its catalog: DeepSeek-V4-Flash 0731 and GLM-5.3.
Snowflake has added dynamic model routing to its Cortex AI Gateway, giving enterprise customers an automated way to reduce AI inference spending without managing model selection themselves. Instead of assigning a fixed model to each task, customers can now select an “auto” setting that routes each query to whichever model offers the best combination of quality and cost.
The announcement comes as enterprise AI budgets face mounting pressure. Worldwide AI spending is projected to hit $2.59 trillion in 2026, up 47% year over year, and 79% of enterprises have already overrun their AI budgets. The FinOps Foundation found that 73% of organizations reported AI costs exceeded original projections, and the share of FinOps teams managing AI spend jumped from 31% in 2024 to 98% in 2026.
How the Routing Works
Snowflake’s system operates through two mechanisms. The first is an advisor pattern, where a smaller model attempts a task and escalates to a larger model only if it cannot complete the work. The second is a classifier trained on past queries that automatically directs straightforward requests to simpler, cheaper models.
The system also builds agent memory into each request. As an agent handles repeated queries, its memory updates and feeds into future interactions, so the system does not re-solve the same problem from scratch. Customers who want to pin a specific model can still do so; auto-routing is optional.
In benchmark testing, agents using dynamic routing built a dbt pipeline with up to 3x greater token efficiency compared to a frontier-model-only approach, while maintaining equivalent output quality. In a separate test, engineering teams completed the same volume of pull requests with 25% greater token efficiency.
New Models and Benchmarks
Snowflake also expanded its model catalog with DeepSeek-V4-Flash 0731 and GLM-5.3. Internal evaluations showed DeepSeek-V4-Flash scoring 74.4% on data engineering tasks. GLM-5.2 scored 62.8% while using fewer tokens than other models tested.
Gartner estimates global token consumption will reach approximately 300 trillion tokens per day by the end of 2026, a figure that makes per-token cost control a material business concern for any organization running AI at scale.
Governance Built In
Cortex AI Gateway gives administrators visibility into token usage and costs across AI apps and agents. Snowflake’s CoCo feature extends those controls through the company’s existing role-based access and tagging framework. Administrators can set default models, attribute usage to specific teams or cost centers, establish per-user quotas, and receive alerts as consumption approaches defined limits.
Dynamic routing also gives customers control over which models and providers are available to their users. Baris Gultekin, Snowflake’s VP of AI, told VentureBeat: “For high quality, enterprise grade agents to be built, it’s crucial to get the context and the governance right. Context, trust and model choice all go hand in hand.”
Industry analyst Sanjeev Mohan of SanjMo noted that the key differentiator is not the routing itself. “Snowflake isn’t really selling routing,” Mohan said. “It’s selling routing that never leaves the governed data boundary, with access controls, tagging, and cost attribution already attached.”
The Business Case
CEO Sridhar Ramaswamy framed the announcement around what Snowflake calls “intelligence efficiency,” a measure of how effectively organizations convert compute, models, data, and context into business results. “The question is no longer how much AI they are using, but whether that AI is translating into meaningful business value,” Ramaswamy said.
Snowflake serves more than 13,900 customers globally. The Cortex AI Gateway updates routing decisions automatically as model performance and pricing change, meaning customers do not need to rebuild workflows each time a new model enters the market.
This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.