Description
Snowflake Cortex AI is Snowflake’s built-in suite of AI and machine learning services that run directly inside the Snowflake AI Data Cloud, without moving data to an external AI platform. It gives SQL and Python users access to hosted large language models from Anthropic, OpenAI, Meta, Mistral, Google, and others through AI Functions like AI_COMPLETE, AI_CLASSIFY, AI_EXTRACT, and AI_EMBED. Cortex Analyst converts natural language questions into SQL queries, Cortex Search handles retrieval-augmented generation over unstructured data, and Cortex Agents orchestrate multi-step workflows across both. Cortex Code, announced in February 2026, adds an AI coding agent for data engineering tasks inside Snowflake Notebooks and local development environments. Because everything runs inside Snowflake’s existing security perimeter, Cortex AI inherits the platform’s role-based access control and governance policies rather than requiring separate infrastructure. It suits organizations already running Snowflake that want LLM-powered analytics without exporting data to a third-party AI service.
Key Features
- AI Functions — SQL-callable functions like AI_COMPLETE, AI_CLASSIFY, AI_EXTRACT, and AI_EMBED for text and image tasks.
- Cortex Analyst — converts natural language questions into SQL queries over your data.
- Cortex Search — retrieval-augmented generation service with managed embedding and indexing.
- Cortex Agents — orchestrates multi-step workflows across Cortex Analyst, Cortex Search, and LLMs.
- Snowflake Intelligence — conversational interface for business users to query data in natural language.
- Cortex Code — AI coding agent for data engineering, announced February 2026.
- Multi-model access — hosted models from Anthropic, OpenAI, Meta, Mistral, and Google, selectable per task.
How It Works
Cortex AI runs inside an existing Snowflake account, so there’s no separate signup or infrastructure to provision. Users call AI Functions directly from SQL or Python, for example running AI_COMPLETE on a text column to summarize or classify rows without exporting data. Cortex Analyst and Cortex Search can be set up over specific tables or documents to answer natural language questions or power retrieval-augmented applications. Cortex Agents chain these services together for multi-step tasks, such as retrieving a document, analyzing it, and generating a recommendation. Snowflake Intelligence wraps this in a conversational interface so business users can ask questions without writing SQL. Cortex Code brings AI-assisted coding into Snowflake Notebooks and local development environments for data engineering work.
Technical Architecture & Overview
- Core Engine: Not one proprietary model; hosts models from Anthropic, OpenAI, Meta, Mistral, Google, and Snowflake’s own Arctic model, selected per function call.
- Deployment: Fully managed and serverless inside the Snowflake AI Data Cloud on AWS, Azure, or Google Cloud; no separate AI infrastructure to provision.
- API Surface: SQL-callable AI Functions, a Python API, and the Cortex REST API, all billed per million tokens processed.
- Known Limits: AI Credit pricing is billed per million tokens (Cortex REST API, AI Functions) or per GB/month indexed (Cortex Search), not a flat per-seat fee; costs scale directly with usage and model choice.
Pros & Cons
| Pros | Cons |
|---|---|
| Runs inside Snowflake’s existing security perimeter, inheriting RBAC and governance policies rather than requiring a separate AI platform | Only usable within the Snowflake ecosystem; not a fit for teams without a Snowflake warehouse |
| Access to hosted models from Anthropic, OpenAI, Meta, Mistral, and Google without separate API keys or contracts | No flat consumer or starter pricing tier; costs are entirely usage-based and can be hard to predict for new users |
| AI Credit pricing is flat across Snowflake editions (Standard, Enterprise, Business Critical), so AI costs don’t rise with edition tier | Cortex Search bills serving compute continuously while a service is resumed, even during idle periods |
| Cortex Agents and Snowflake Intelligence orchestrate multi-step workflows without custom pipeline code | Cortex Fine-tuning and the standalone Cortex Analyst API still use legacy Platform Credit pricing rather than the newer AI Credit model |
| No per-seat fees; cost scales directly with token and compute usage | Regional routing costs 10% more per AI Credit ($2.20) than global routing ($2.00), a tradeoff for data residency needs |
Pricing
| Item | Price |
|---|---|
| AI Credits (Global routing) | $2.00 per AI Credit (flat across all Snowflake editions) |
| AI Credits (Regional routing) | $2.20 per AI Credit (for data residency or compliance requirements) |
| Platform Credits (Enterprise Edition) | $3.00 per credit (non-AI compute, storage, warehouse usage) |
Platform Availability
Web (Snowsight) | SQL | Python | REST API | AWS | Azure | Google Cloud
Best For
Enterprise data teams | Data analysts | ML engineers | Regulated industries | Snowflake customers
Frequently Asked Questions
Is there a free plan for Snowflake Cortex AI?
No. Cortex AI has no flat consumer or free tier. It uses AI Credit pricing billed per million tokens processed or per GB of indexed data, on top of an existing Snowflake account, per Snowflake’s official documentation.
How much does an AI Credit cost?
An AI Credit costs $2.00 with global routing or $2.20 with regional routing, and this price is flat across all Snowflake editions (Standard, Enterprise, Business Critical), according to Snowflake’s official AI pricing page.
Which AI models are available through Cortex AI?
Cortex AI hosts models from Anthropic, OpenAI, Meta, Mistral, and Google, selectable per function call, alongside Snowflake’s own Arctic model, without requiring separate vendor contracts.
How does Snowflake Cortex AI differ from a standalone database tool like DBeaver?
Cortex AI runs generative AI functions and agents natively inside the Snowflake warehouse using AI Credit billing, while a client tool like DBeaver connects to a database from a desktop app and adds AI features as an assistant layer rather than in-warehouse functions.
What is Cortex Code?
Cortex Code is Snowflake’s AI coding agent, announced February 2026, that brings context-aware coding assistance for data engineering tasks into Snowflake Notebooks and local development environments.
