LandingAI Releases Agentic Document Extraction Gen2 with DPT-3 Pro and DPT-3 Verity
LandingAI rebuilt its document extraction stack around agentic primitives. Gen1 code breaks on Gen2 endpoints.
Why it matters
LandingAI's Gen2 migration signals a shift in how document AI vendors architect their stacks — moving from chunk-based extraction to document-tree reasoning and per-word grounding. Practitioners using Gen1 face a breaking API change; the new billing model (output characters vs. flat pages) changes unit economics.
The key facts
12 to knowAgentic Document Extraction Gen2 released
DPT-3 model family: DPT-3 Pro (line-level grounding), DPT-3 Verity (word-level grounding with confidence scores)
Architecture shift: chunks retired in favor of document/page/block tree structure
Billing model changed from flat pages to output characters
Gen1 code incompatible with Gen2 endpoints (breaking API change)
Published: September 9, 2026
LandingAI ships Agentic Document Extraction Gen2
Built on DPT-3 model family (Pro, Verity variants)
Architectural shift: chunks → document/page/block tree
DPT-3 Pro grounds to line-level; DPT-3 Verity grounds to word with confidence scores
Billing model change: output characters instead of flat pages
Breaking change: Gen1 code will not run on Gen2 endpoints
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: LandingAI has shipped Agentic Document Extraction Gen2, a rebuild of its document stack on the DPT-3 model family. Chunks are retired in favor of a document, page and block tree. DPT-3 Pro grounds to the line, DPT-3 Verity grounds to the word with a confidence score, and Parse billing now counts…