Snowflake Cortex has become one of the most powerful AI toolkits available natively inside a data platform. In 2026, the suite now includes full LLM inference, document AI, vector search, and a growing library of ML functions — all running directly on your Snowflake data without moving it to an external model service.
What's New in Cortex
The most significant additions this year are:
- CORTEX.COMPLETE — Run LLM completions (Llama 3, Mistral, Arctic) against any string column in a SELECT statement.
- CORTEX.SEARCH — Hybrid semantic + keyword search over Snowflake tables, powered by built-in vector embeddings.
- Document AI — Extract structured data from PDFs and images stored in Snowflake stages using a point-and-click interface.
- Cortex Analyst — Natural language to SQL, embedded directly into apps via the REST API.
What's Production-Ready?
From our engagements, CORTEX.COMPLETE and CORTEX.EMBED_TEXT are the most stable for production use. Document AI is GA and handling enterprise workloads reliably. Cortex Analyst is still best suited for internal tools rather than customer-facing applications.
Our Recommendation
If you haven't yet evaluated Cortex for your Snowflake estate, start with CORTEX.COMPLETE on your existing text data. The fact that it operates on data already in Snowflake — with no API keys, no data movement, and governed by your existing Snowflake RBAC — is a genuine competitive advantage over external LLM calls.
Contact us if you'd like a free Cortex readiness assessment for your Snowflake environment.