European Investor
18 Aug 2026, 17:11
Snowflake Expands Cortex AI With Dynamic Model Routing to Cut Enterprise AI Costs
Snowflake (NYSE: SNOW) announced new artificial intelligence capabilities on Tuesday aimed at helping enterprises reduce the cost of deploying AI applications and agents at scale.
The company is introducing dynamic model routing within Cortex AI Gateway, allowing AI workloads to be automatically directed to different models based on factors including quality, speed, customer preferences and cost. Snowflake is also expanding its model portfolio with DeepSeek-V4-Flash 0731 and GLM-5.3.
Dynamic Routing Targets Rising AI Inference Costs
Instead of using an expensive frontier AI model for every request, Cortex AI Gateway can determine which model provides an appropriate balance between performance and cost. Simpler or repetitive workloads can be routed toward more efficient models, while tasks requiring deeper reasoning can be directed toward frontier models.
The capability will be integrated with Snowflake's flagship AI products, including Snowflake CoCo and Snowflake CoWork, and will also be available to third-party AI agents connected through Cortex AI Gateway.
The strategy addresses an increasingly important issue for enterprises adopting generative and agentic AI: inference costs can rise rapidly as AI workloads move from experimentation into large-scale production.
Snowflake said internal testing showed meaningful efficiency improvements. In one evaluation, agents using dynamic model routing to build a dbt pipeline achieved as much as 3x greater token efficiency than a frontier-model-only approach while maintaining comparable quality. Another test showed engineering teams completing the same number of pull requests with 25% greater token efficiency.
Snowflake Expands Open AI Model Options
Snowflake is also adding DeepSeek-V4-Flash 0731 and GLM-5.3 to its AI ecosystem, expanding a model portfolio that already includes models from providers such as Anthropic, OpenAI, Google, SpaceXAI, Meta and Mistral.
Snowflake's own testing highlights the potential economics behind the strategy. DeepSeek-V4-Flash scored 74.4% on data-engineering tasks in the company's evaluation, outperforming the proprietary models included in that test, while GLM-5.2 scored 62.8% and consumed fewer tokens than any other model evaluated.
The company is pairing broader model access with tools allowing enterprises to monitor token usage and costs, allocate AI spending across teams, establish quotas and spending limits, and control which models employees and AI agents can access.
For Snowflake, the initiative strengthens its position at the intersection of enterprise data infrastructure and AI. Rather than competing solely through access to individual AI models, Snowflake is positioning Cortex AI Gateway as an orchestration and governance layer that can determine which model should handle each workload while keeping enterprise data within Snowflake's governed environment.
Snowflake (NYSE: SNOW) announced new artificial intelligence capabilities on Tuesday aimed at helping enterprises reduce the cost of deploying AI applications and agents at scale.
The company is introducing dynamic model routing within Cortex AI Gateway, allowing AI workloads to be automatically directed to different models based on factors including quality, speed, customer preferences and cost. Snowflake is also expanding its model portfolio with DeepSeek-V4-Flash 0731 and GLM-5.3.
Dynamic Routing Targets Rising AI Inference Costs
Instead of using an expensive frontier AI model for every request, Cortex AI Gateway can determine which model provides an appropriate balance between performance and cost. Simpler or repetitive workloads can be routed toward more efficient models, while tasks requiring deeper reasoning can be directed toward frontier models.
The capability will be integrated with Snowflake's flagship AI products, including Snowflake CoCo and Snowflake CoWork, and will also be available to third-party AI agents connected through Cortex AI Gateway.
The strategy addresses an increasingly important issue for enterprises adopting generative and agentic AI: inference costs can rise rapidly as AI workloads move from experimentation into large-scale production.
Snowflake said internal testing showed meaningful efficiency improvements. In one evaluation, agents using dynamic model routing to build a dbt pipeline achieved as much as 3x greater token efficiency than a frontier-model-only approach while maintaining comparable quality. Another test showed engineering teams completing the same number of pull requests with 25% greater token efficiency.
Snowflake Expands Open AI Model Options
Snowflake is also adding DeepSeek-V4-Flash 0731 and GLM-5.3 to its AI ecosystem, expanding a model portfolio that already includes models from providers such as Anthropic, OpenAI, Google, SpaceXAI, Meta and Mistral.
Snowflake's own testing highlights the potential economics behind the strategy. DeepSeek-V4-Flash scored 74.4% on data-engineering tasks in the company's evaluation, outperforming the proprietary models included in that test, while GLM-5.2 scored 62.8% and consumed fewer tokens than any other model evaluated.
The company is pairing broader model access with tools allowing enterprises to monitor token usage and costs, allocate AI spending across teams, establish quotas and spending limits, and control which models employees and AI agents can access.
For Snowflake, the initiative strengthens its position at the intersection of enterprise data infrastructure and AI. Rather than competing solely through access to individual AI models, Snowflake is positioning Cortex AI Gateway as an orchestration and governance layer that can determine which model should handle each workload while keeping enterprise data within Snowflake's governed environment.
