Snowflake, a company specializing in the AI Data Cloud, has announced the integration of dynamic model routing into Cortex AI Gateway and its leading AI products. At the same time, the company has expanded its offerings with some of the most prominent open models available in the market.
With these innovations, Snowflake aims to help businesses optimize AI spending and improve what it calls “intelligence efficiency.” This metric seeks to measure how effectively an organization can transform resources like processing power, AI models, data, and business context into business value outcomes.
The new features build upon Cortex AI Gateway, launched by Snowflake in July 2026 as a unified platform designed to manage agent connections, intelligently route requests, and optimize the use of AI resources.
The growth of AI applications and agents used in production environments is presenting new challenges for organizations. Using a single model for all tasks can significantly increase costs, while managing a growing number of models forces development teams to spend more time evaluating, configuring, and maintaining them.
To address this, Snowflake is focusing on automating the selection of the most appropriate model for each request, while expanding the options available to its customers. This allows organizations to combine open-source models and proprietary solutions based on the needs of each task, seeking a more efficient and cost-effective use of AI.
With the addition of dynamic model routing, Cortex AI Gateway can automatically select the specific model that offers the best balance between quality and cost for each specific task. Dynamic model routing is also integrated into Snowflake’s flagship AI products, including Snowflake CoCo and Snowflake CoWork, and is available for third-party AI agents utilizing Cortex AI Gateway. This new functionality redirects repetitive or low-complexity tasks to more efficient models, while those requiring higher reasoning are routed to boundary models. This helps clients reduce unnecessary inference costs without needing to manage model selection themselves. Snowflake will also expand access for customers to leading open models such as DeepSeek-V4-Flash 07311 and GLM-5.33, through Snowflake Cortex AI. This complements Snowflake’s extensive model library, giving customers more options to balance model quality and cost, all while keeping governed data secure within Snowflake.
Snowflake’s latest innovations give companies greater control over AI cost efficiency as their AI usage scales, helping them improve intelligence efficiency by assigning each task to the right model and reducing unnecessary expenses.
“Companies are increasingly stringent about AI costs and profitability. The question is no longer how much AI they’re using, but whether that AI delivers significant business value,” says Sridhar Ramaswamy, Snowflake CEO. “Achieving intelligence efficiency requires the flexibility to use the best model for each task as the landscape evolves. Snowflake’s role is to absorb that complexity so customers can focus on outcomes while we optimize model selection in the background.”
“Organizations are overwhelmed by the sheer number of model options, but the real challenge isn’t choosing the model — it’s the operational burden of selecting the right one at scale. Snowflake’s dynamic model routing directly addresses this gap,” states Sanjeev Mohan, director and founder of SanjMo. “By automating intelligent model selection within Cortex AI Gateway, Snowflake removes a real friction point currently hindering enterprise AI deployment. The ability to align workload complexity with model cost, without rebuilding infrastructure each time a new model emerges, is exactly the kind of efficiency companies need to move from experimentation to scaled AI.”
Cortex AI Gateway optimizes AI cost through model selection
Dynamic model routing in Cortex AI Gateway is designed to help organizations use boundary models efficiently—only when they significantly improve outcomes in Snowflake CoCo, Snowflake CoWork, and third-party AI agents. Beyond automatically selecting a model for each request, dynamic routing allows clients to control which models and providers are available to their users. This is especially important for global organizations managing regional model availability or for regulated industries with specific compliance requirements. As model performance and pricing change, Cortex AI Gateway can update routing decisions within Snowflake’s AI products, such as Snowflake CoCo and Snowflake CoWork, preventing the need for application or agent redesigns. This allows companies to capitalize on new model options as they become available, while Snowflake manages the complexity of model selection and optimization behind the scenes.
Internal testing indicates that combining open and proprietary models for various tasks can deliver comparable quality while significantly improving token efficiency. In one evaluation, agents utilizing Cortex AI Gateway’s dynamic model routing created data transformation workflows with dbt consuming up to 3 times fewer tokens than when using only boundary models, maintaining the same quality. In another test, engineering teams completed the same number of pull requests with 25% higher token efficiency⁴.
Snowflake introduces new open models for governed enterprise data
DeepSeek-V4-Flash 0731 and GLM-5.3 are the latest models added to Snowflake’s expanding portfolio available on Cortex AI, Snowflake CoCo, and Snowflake CoWork. Along with models from providers like Anthropic, OpenAI, Google, SpaceXAI, Meta, and Mistral, this portfolio offers customers greater flexibility to choose the optimal mix of performance and cost for each workload.
Rather than viewing each open model release as an independent integration, Snowflake is creating an ecosystem designed to stay current with the fast-evolving open model landscape. Snowflake continuously evaluates and optimizes how these models are served and used within its AI offerings, helping clients leverage open-source advancements without constantly redesigning their applications or infrastructure. Consequently, users can benefit from newer, more capable models while maintaining consistent access controls and governance—whether for open or proprietary options.
Snowflake’s AI Research team evaluated DeepSeek-V4-Flash in enterprise-oriented tasks; recent tests show that DeepSeek v4 Flash outperformed leading proprietary models in the evaluation, scoring 74.4% on data engineering tasks. GLM-5.2 also achieved a solid 62.8% performance, using fewer tokens than any other tested model⁵. These results demonstrate that open models are increasingly capable of supporting mission-critical tasks at a lower cost.
Snowflake helps organizations improve AI efficiency at scale
With deploying multiple models across a growing number of AI applications and agents, enhancing efficiency requires visibility and control over resource consumption. Cortex AI Gateway provides administrators with transparency into token usage and costs, while enabling organizations to set spending limits across all AI applications and agents. Snowflake CoCo extends these controls through its existing tagging framework and role-based access, allowing admins to set default models, attribute consumption to teams or cost centers, establish user quotas, and receive notifications as consumption approaches limits.
By combining dynamic model routing with a broader model selection, these capabilities give companies greater control over costs and profitability. Organizations can automatically adapt workloads to appropriate models, take advantage of new options as they arise, and manage resource consumption across the enterprise. This approach helps improve AI efficiency while maintaining Snowflake’s trusted governance and keeping proprietary data and business context secure.
