As enterprises move AI agents from experimentation into production, the cost of choosing the wrong model for routine tasks can quickly become significant. Snowflake is addressing that problem with dynamic model routing in Cortex AI Gateway, allowing workloads to be matched automatically with models based on task complexity, quality requirements and cost. The company is also expanding access to open models, giving enterprises more options as AI workloads scale.

Snowflake is expanding its enterprise AI infrastructure with dynamic model routing designed to help businesses control inference costs without forcing developers to manually select a model for every AI task.

The capability is being introduced within Snowflake Cortex AI Gateway and integrated into Snowflake's AI products, including Snowflake CoCo and Snowflake CoWork. It is also available to third-party AI agents connected through Cortex AI Gateway.

The underlying idea is straightforward: not every AI request needs a frontier model.

A repetitive classification task, simple data transformation or routine workflow may be handled effectively by a less expensive model. A complex reasoning problem, coding task or sophisticated agent workflow may justify a more capable—and potentially more expensive—model.

Snowflake's dynamic routing system is designed to make that decision automatically.

The move comes as enterprise AI deployments become increasingly multi-model. Organizations can now choose from proprietary systems offered by companies such as OpenAI, Anthropic and Google, alongside rapidly evolving open models from providers such as DeepSeek, Meta and Mistral.

That choice creates flexibility, but it also creates operational complexity.

Model Choice Becomes an Infrastructure Problem

When organizations run a handful of AI experiments, model selection is relatively straightforward. Developers can test several models and choose one based on quality and price.

Production AI is different.

An enterprise may have hundreds or thousands of agents making requests across customer service, analytics, software development, marketing and internal operations. Model pricing and capabilities can also change quickly.

Using the most capable model for every request can increase costs unnecessarily. But optimizing manually creates another problem: engineering teams must continually monitor model performance, pricing, availability and compliance requirements.

Snowflake is positioning Cortex AI Gateway as an abstraction layer that handles some of that complexity.

Its dynamic routing capability can direct lower-complexity work toward more efficient models while sending tasks requiring deeper reasoning to frontier systems.

The approach effectively turns model selection into an infrastructure function rather than a decision developers must repeatedly build into individual applications.

Snowflake's "Intelligence Efficiency" Strategy

Snowflake describes this approach as intelligence efficiency—a measure of how effectively organizations convert compute, models, data and context into business value.

That framing is significant because enterprise AI economics are becoming more complicated than simply measuring the number of tokens consumed.

A less expensive model may be the better choice if it produces an acceptable result with significantly lower inference costs. Conversely, a more powerful model can make economic sense if better reasoning reduces errors, rework or downstream human intervention.

Snowflake says internal testing demonstrated the potential impact.

In one evaluation, agents using dynamic routing to build a dbt pipeline achieved up to three times greater token efficiency than a frontier-model-only approach while maintaining comparable quality.

In another test, engineering teams completed the same number of pull requests with 25% greater token efficiency.

Those are Snowflake's internal results rather than independently verified industry benchmarks, so they should be viewed as directional rather than universal performance expectations.

Still, they illustrate the central economic argument: model quality and model cost do not necessarily need to move together.

Open Models Expand the Enterprise AI Mix

Snowflake is also adding more open models to its Cortex AI environment, including DeepSeek-V4-Flash 0731 and GLM-5.3.

The company already provides access to models from several major providers, including Anthropic, OpenAI, Google, xAI, Meta and Mistral.

For enterprises, the importance of that growing model catalog extends beyond having more choices.

Open and proprietary models can have different cost structures, capabilities, licensing conditions, deployment characteristics and regional availability.

A multinational enterprise, for example, may need to consider where a model is available and whether its use complies with internal data policies or regional requirements.

Snowflake says Cortex AI Gateway allows customers to control which models and providers are available to users.

That governance layer could become increasingly important as enterprises introduce AI agents into sensitive workflows.

Governance Meets Multi-Model AI

The development also illustrates why model routing is becoming closely connected with enterprise AI governance.

A company may want employees to have access to multiple models, but that does not mean every employee or application should have unrestricted access to every provider.

Centralized controls can help organizations define approved models, manage access and account for regional requirements.

Snowflake's approach is particularly relevant to enterprises that already keep data and AI workflows inside its platform.

Rather than requiring developers to integrate every new model independently, Snowflake is attempting to create a governed environment in which models can be added and substituted behind the application layer.

That could reduce the infrastructure burden associated with the rapid pace of model releases.

Competition Is Moving Up the AI Stack

Snowflake is entering a competitive enterprise AI infrastructure market that includes Microsoft Azure, Amazon Web Services, Google Cloud and Databricks, among others.

Cloud providers increasingly offer model catalogs, routing mechanisms, agent infrastructure and governance tools. Data platforms are also moving toward becoming orchestration layers for enterprise AI.

Snowflake's advantage is its existing position around enterprise data.

The company is betting that customers want model flexibility without moving governed data between multiple environments or rebuilding applications whenever a better model becomes available.

That strategy also aligns with the broader shift from single-model AI applications toward multi-model architectures.

The winning platform may not necessarily be the one with the single strongest model. It could be the one that helps enterprises choose, govern and operate multiple models efficiently.

Market Landscape

Enterprise AI is entering a multi-model era.

Organizations increasingly use different models for different workloads rather than standardizing on a single provider. That creates opportunities for model gateways and orchestration platforms that can manage routing, access, security and cost.

Snowflake's Cortex AI Gateway competes indirectly with capabilities emerging across Microsoft Azure AI, Amazon Bedrock, Google Cloud Vertex AI and Databricks Mosaic AI.

The competitive differentiator is therefore shifting from model access alone to model optimization and governance.

For enterprise marketing teams, the implications extend beyond IT. AI-powered content generation, customer analytics, campaign optimization and marketing agents can generate large volumes of model requests. Automatically routing those requests to appropriate models could help organizations control costs while maintaining acceptable output quality.

Strategic Outlook

The economics of enterprise AI will increasingly depend on orchestration rather than simply model performance.

As models become more numerous and specialized, organizations will need infrastructure that can evaluate quality, cost, latency, availability and governance requirements simultaneously.

Dynamic routing is one response to that problem.

Snowflake's strategy suggests the future enterprise AI stack may look less like a collection of applications tied to individual models and more like a governed intelligence layer capable of switching models underneath those applications.

For enterprise marketing teams, that could eventually mean AI agents that use inexpensive models for routine segmentation or content tasks and more sophisticated reasoning models only when a campaign or customer decision warrants the additional cost.

The broader lesson is that AI efficiency may increasingly come from using the right model at the right moment, rather than simply finding the most powerful model available.

Top Insights

• Snowflake's dynamic model routing automatically matches AI workloads with appropriate models, helping enterprises reduce inference costs without manually managing model selection.
• Cortex AI Gateway gives enterprises greater control over multi-model environments, including provider access, regional availability and governance requirements across AI applications.
• Snowflake's internal tests reported up to three-times greater token efficiency for certain agent workflows, though independent benchmarking is needed to validate broader results.
• New DeepSeek and GLM models expand Snowflake's open-model options, giving enterprises more flexibility to balance performance, cost and governance.
• Model routing positions Snowflake against cloud and data-platform competitors as enterprise AI shifts toward multi-model architectures and centralized orchestration.

Get in touch with our MarTech Experts