/
New feature

Databricks Document Intelligence

Databricks Document Intelligence is a Databricks ai / ml capability, introduced April 2026.

Databricks' umbrella name for its intelligent document processing stack - a set of research-built AI Functions (ai_parse_document, ai_extract, ai_classify, and ai_prep_search in Beta) that parse, extract, and classify PDFs, images, Word files, and slides at scale in SQL, through REST APIs, or in the no-code Agent Bricks UI.

  • It was born out of a benchmark embarrassment - Databricks found that even highly capable frontier agents scored below 50% accuracy on real enterprise document tasks, and concluded the bottleneck was reading, not reasoning.
  • Swapping nothing but the document layer helps - pre-processing treasury bond documents with ai_parse_document gave a 16% average gain across every agent framework Databricks tested, with the agents themselves left untouched.
  • Its hardest-mode extractor manages a team - ai_extract Precision Mode reasons in stages, spawns subagents to extract in parallel, and merges their results into one output.

Limitations: Its parsing step, ai_parse_document, is limited to documents of at most 500 pages and 100 MB - a longer document fails immediately unless you pass a page range that stays within the limit.

Open in REbricked →
Category
AI / ML
Introduced
April 2026
Announced at
Announced April 16, 2026, the same day ai_parse_document went GA
Also known as
Document Intelligence, Intelligent document processing, IDP
Verified
2026-09-27

Sources

Related in AI / ML