New functionality in Cortex AI Gateway automatically selects the right AI model for each task. In Snowflake’s own testing, token efficiency increased by up to three times while maintaining quality.
Snowflake, the company behind the AI Data Cloud, is launching dynamic model routing, a new feature in Cortex AI Gateway and Snowflake’s flagship AI products that automatically selects the AI model that provides the best balance of quality, speed, and cost for a given task. The goal is to help enterprises reduce unnecessary AI costs as the use of AI applications and agents increases.
Instead of using the same and often the most advanced and expensive AI model for all tasks, dynamic model control in Cortex AI Gateway Direct simpler and repetitive tasks to more efficient models. Tasks that require more advanced analysis and reasoning can be allocated to more advanced and powerful models. 
In Snowflake’s internal testing, the approach has yielded clear efficiency gains. AI agents using dynamic model control with Cortex AI Gateway built a DBT pipeline with up to three times higher token efficiency while maintaining quality, compared to using a frontier model alone. In a separate test, development teams executed the same number of pull requests with 25 percent higher token efficiency. *
–Companies are becoming more and more careful about the economics of AI. The question is no longer how much AI they use, but whether that AI is being translated into meaningful business value, says Sridhar Ramaswamy, CEO of Snowflake. Achieving efficiency with AI requires the flexibility to use the model best suited for each individual task as the landscape evolves. Snowflake's role is to manage that complexity so customers can focus on results while we optimize model selection in the background.
Snowflake simultaneously expanding the range of open AI models in Snowflake Cortex AI. DeepSeek-V4-Flash 0731 becomes available in private preview and GLM-5.3 becomes available in private preview soon. They complement models from Anthropic, OpenAI, Google, SpaceXAI and Mistral, among others.
In addition, new functions are also being launched to follow up and control AI usage and costs, set spending limits and quotas, and distribute AI costs between, for example, different departments and cost centers.
Dynamic model routing integrates into Snowflake's AI products Snowflake CoCo and Snowflake CoWork and can also be used by third-party AI agents via the Cortex AI Gateway.
The news in brief:
- Automatic model control based on quality, speed and cost, among other things
- Up to 3x higher token efficiency in Snowflake internal testing*
- New open models, including DeepSeek-V4-Flash 0731 and GLM-5.3
- New tools to track, allocate and limit companies' AI costs
- Data and AI usage are managed within Snowflake's security and governance framework
About Snowflake
Snowflake is the platform for the AI era, making it easy for businesses to innovate faster and get more value from data. More than 13,900 customers worldwide, including hundreds of the world’s largest enterprises, use Snowflake’s AI Data Cloud to securely build, consume, and share data, apps, and AI. With Snowflake, data and AI are transformative for everyone.
Swedish 
Danish 
Norwegian 
Finnish 
English 
German 
Dutch 