The new capability automatically selects AI models based on cost, performance, and latency policies, with Snowflake claiming up to 3x greater token efficiency in internal tests.