Comparison · Snowflake

Snowflake Cortex alternative: seven options compared

Cortex is the shortest path to AI if your data already sits in Snowflake. The question is whether the AI Credit bill and the in-warehouse boundary fit your workload better than an external API.

Short answer. Use Cortex when data gravity and Snowflake governance dominate. Use Plugsky for a flat managed API; Databricks for Unity Catalog; Microsoft Foundry, AWS Bedrock or Google's platform for hyperscaler governance; OpenRouter for breadth; Together for GPUs. Compare the AI Credit rate, data path and model catalogue.

Cortex functions vs an external API

Cortex is a suite, not one endpoint: AI Functions such as AI_COMPLETE, AI_CLASSIFY, AI_EMBED, AI_EXTRACT, AI_SUMMARIZE and AI_TRANSLATE, plus Cortex Agents, Analyst, Search, fine-tuning and the Cortex AI Gateway. AI Functions run as SQL and are optimized for throughput; the REST and gateway APIs are better for interactive latency. Models run inside Snowflake's perimeter without moving data to a separate AI vendor.

DimensionCortex AI FunctionsCortex AI GatewayExternal AI API
Where inference runsInside SnowflakeInside SnowflakeVendor's infrastructure
Data movementNone to an external modelNone to an external modelPrompts and responses leave your boundary
Billing unitAI Credits per million tokensAI Credits per tokenPer token, per second, or flat
Model choiceSnowflake-supported catalogueSnowflake-supported catalogueBroad, vendor-dependent
GovernanceSnowflake roles and privilegesUSAGE/MONITOR plus tracesVendor RBAC, SSO, audit
Latency shapeOptimized for batch throughputREST-style interactiveVendor-dependent

What Cortex is good at, and what it is not

Cortex is excellent when data is already in Snowflake and the job is SQL-shaped: classify tickets, extract entities, summarize a table, embed a corpus, redact PII, translate a column. Data does not leave the perimeter, roles are familiar, and results land in a table. Cortex Search adds retrieval, and Cortex Agents orchestrate multi-step workflows with cost attribution.

Cortex is not a general-purpose multi-provider gateway. The catalogue is whatever Snowflake supports, and the price is an AI Credit rate rather than a provider's per-token list. AI Functions optimize throughput; Snowflake recommends the REST API for interactive latency. Payload logging is off by default but can be turned on, making the trace table sensitive. Availability varies by region, and the 60-day deprecation notice for GA models is a planning constraint.

The seven alternatives

OptionBest forData movementBillingWatch out
PlugskyFlat multi-model API with private deploymentAPI call; VPC/on-prem optionalFlat $8–$120/monthNot a warehouse; confirm model fit
Databricks Unity GatewayUnity Catalog governanceInto the lakehouseDBUs per token/hourDBU math; platform coupling
Microsoft FoundryAzure-native agents and modelsAzure tenantAzure consumptionAzure governance required
AWS BedrockAWS-native models and GuardrailsAWS accountAWS consumptionRegion and model availability
Google Gemini Enterprise Agent PlatformGoogle Cloud models and Model GardenGoogle Cloud projectGoogle Cloud consumptionPlatform-specific tooling
OpenRouterCatalogue breadth and fallbackProvider-dependentList price + credit feeProvider variance
Together AIOpen models and dedicated GPUsTogether infrastructurePer token, per video, or GPUData leaves your boundary

Plugsky is the flat bill with 39 models and private deployment. Databricks is the closest architectural peer if your data is in a lakehouse; Unity Gateway governs models, agents and MCP tools through Unity Catalog, with inference tables and DBU pricing. Microsoft Foundry, AWS Bedrock and Google's platform are the hyperscaler options. OpenRouter is breadth; Together is open models and GPUs.

Cost: AI Credits vs per-token

Snowflake prices AI features in AI Credits: $2.00 per credit for global routing and $2.20 for regional. AI Functions, Agents, Search, the gateway and the REST API consume credits at model-specific rates; the rate is in the consumption table, not a single per-token number. Platform credits for warehouses and storage are separate, and capacity discounts do not apply to AI Credits.

Convert the model's AI Credit rate to dollars per million tokens: credits per million × $2.00 (or $2.20 regional). If a model consumes 5 credits per million, the AI cost is $10 per million before warehouse compute — an illustration, not a Snowflake rate. Add warehouse time, storage and Cortex Search serving per GB per month. An external quote usually includes only tokens, so add your compute and egress.

Cost componentCortexExternal API
Model inferenceAI Credits per token, model-specificPer token, per second or flat
ComputeVirtual warehouse creditsYour app or batch compute
RetrievalCortex Search serving + embeddingVector database + embedding
Data movementUsually noneEgress and API transfer
DiscountsACV-based AI Credit discountsProvider or plan discounts

Data movement and security

The strongest argument for Cortex is the data path. Snowflake says models provided through its AI features are deployed within the Snowflake perimeter, customer data is not available to other customers or model developers, and it does not train on customer data. That reduces subprocessors and egress, but not the need for access control, payload-logging decisions and retention rules. Cortex AI Gateway logs metadata by default; payload capture is opt-in.

An external API can be the better security choice when you need a specific model, a private VPC or on-premises deployment, or a control set your auditors accept. The trade is data movement and a larger subprocessor list. Ask both sides where inference runs, where prompts and logs are stored, who can read them, how long they are kept and what happens on deletion. For regulated data, add the BAA or DPA and the residency map.

Migration

Moving off Cortex is an application refactor. Extract SQL functions into service calls, map models to the target catalogue, preserve prompts and output schemas, and test structured output and tool calling. Move embeddings and re-index; do not assume dimensions or similarity scores transfer. Dual-run a sample, compare quality, latency and cost per completed task, then cut over by workload. Keep Snowflake batch jobs if they are still cheaper.

Moving to Cortex is the reverse: land data, define roles, choose Functions or the gateway, set logging and retention, and test the model on your evals. The AI Credit rate is one input; warehouse compute and engineering time are the others. For the SaaS context, see embedding AI without becoming an infrastructure company, the Databricks alternative and Plugsky pricing.

FAQ

What is Snowflake Cortex?

Cortex is Snowflake's AI suite: AI Functions such as AI_COMPLETE, AI_CLASSIFY, AI_EMBED and AI_EXTRACT, plus Cortex Agents, Cortex Analyst, Cortex Search, fine-tuning, Snowflake Intelligence and the Cortex AI Gateway. It runs models inside Snowflake's governance perimeter.

How is Cortex priced?

AI features use AI Credits: $2.00 per credit for global routing and $2.20 for regional routing. Each model consumes a published number of credits per million tokens or per token; warehouse compute, storage and Cortex Search serving are separate.

What is the best Cortex alternative?

It depends on where the data lives: Databricks Unity Gateway for a lakehouse, Plugsky for a flat managed API, Microsoft Foundry, AWS Bedrock or Google's agent platform for hyperscaler governance, OpenRouter for breadth, Together AI for GPUs.

Does Cortex move my data outside Snowflake?

Snowflake states that models provided through its AI features are deployed within the Snowflake Service perimeter, and customer data is not available to other customers or model developers. Confirm the current terms, region and logging configuration for your account.

How hard is migration off Cortex?

It is an application refactor: extract SQL functions into service calls, map models, preserve prompts and output schemas, re-index embeddings, dual-run and compare cost per completed task. Data-local batch jobs may still be cheaper in Cortex.

Sources

Primary sources checked 20 November 2026. Provider terms, limits and prices change; verify the current version before you commit.

  • Snowflake Cortex overview — describes Cortex Agents, AI Functions, Analyst, Search, fine-tuning and the security and data-privacy principles.
  • Snowflake AI pricing — lists AI Credit prices of $2.00 for global routing and $2.20 for regional routing, and the features that consume them.
  • Snowflake Cortex AI Gateway — documents the single governed endpoint, USAGE/MONITOR privileges, trace logging, payload capture and AI Credit billing.
  • Databricks Unity Gateway — provides the comparison point for Unity Catalog governance, inference tables and DBU-based model serving.

Reviewed 20 November 2026. Snowflake AI Credit rates and Cortex capabilities were checked on 20 November 2026. Model-specific credit factors are in the Snowflake consumption table; confirm current rates and regional availability.