AI Functions

AI Functions is a Databricks ai / ml capability, introduced April 2023.

Built-in Databricks SQL functions that apply LLMs or state-of-the-art research techniques to data stored on Databricks for transformation and analysis, callable from Databricks SQL, notebooks, Lakeflow pipelines, and Workflows - the general-purpose ai_query plus a growing family of task-specific functions like ai_extract, ai_classify, ai_parse_document, ai_search, and ai_transcribe.

• The function the 2023 preview shipped with, ai_generate_text, is now marked deprecated in favour of ai_query - yet its doc page still wears a Public Preview badge, so it managed to be deprecated without ever leaving preview. 🔗
• Maturity is a spectrum, not a badge: ai_query reached GA in June 2026 after three years in preview, while sibling ai_analyze_sentiment still tells you it is in Public Preview and HIPAA compliant in the same breath. 🔗
• The newest members are SQL functions with no SQL in sight: the AI Functions REST API went GA in August 2026 so applications can call ai_parse_document, ai_extract, and ai_classify without a warehouse, and ai_transcribe (Beta) will eat an hour of audio per call but politely refuses video. 🔗

⚠ Limitations: Not available on Classic SQL warehouses; Databricks Runtime 15.4 LTS or above is required (18.2 or above recommended). Beta members carry their own hard caps - ai_transcribe accepts at most 1 hour of audio and a 512 MB file per call and does not transcribe video; ai_search takes up to 10 knowledge sources per call and a 4,000-character instructions argument. 🔗

Category

AI / ML

Introduced

April 2023🔗

Also known as

ai_query, ai_extract, ai_search, ai_transcribe, ai_classify, ai_parse_document, ai_generate_text, AI Functions REST API

Verified

2026-09-20