LandingAI Launches Agentic Document Extraction Gen2 with DPT-3 Pro and DPT-3 Verity
LandingAI has released Agentic Document Extraction (ADE) Gen2, a ground-up rebuild of its document intelligence stack centered on a new model family called DPT-3. While Gen1 treated documents as flat lists of chunks, Gen2 models them as hierarchical trees, bills by the characters returned rather than by the page, and grounds every extracted answer to a specific line or word on the page. LandingAI frames the release around three themes: affordability, agent-ready outputs, and atomic grounding.
Is It Deployable?
Yes. ADE Gen2 is generally available now. Developers can start for free in the ADE playground. Enterprises can run it in US or EU cloud regions, in their own VPC on AWS, Azure, or Google Cloud, inside Snowflake, or on premises—including air-gapped environments.
Two Parsing Models Instead of One
Gen2 splits parsing into two models so workloads can choose the appropriate price-performance point. DPT-3 Verity deterministically transcribes digitally created documents and returns a bounding box and confidence score for every word. It targets high-volume text, tables, and simple form fields. DPT-3 Pro reads page layout before words, detects block types ranging from tables and figures to marginalia and signatures, returns them in reading order, and handles scanned pages, handwriting, non-Latin scripts, and LaTeX math. LandingAI states that DPT-3 Verity costs roughly 40% of the credits required by DPT-3 Pro.
Key Takeaways
- Hierarchical document modeling: Gen2 represents documents as trees rather than flat chunk lists.
- Character-based pricing: Billing is based on returned characters, not page count.
- Atomic grounding: Every answer is traced back to a specific line or word.
- Flexible deployment: Available via US/EU cloud, customer VPCs on AWS, Azure, or Google Cloud, Snowflake, and on-premises including air-gapped setups.
- Two-model parsing strategy: DPT-3 Verity for deterministic, high-volume text and simple forms; DPT-3 Pro for complex layouts, scans, handwriting, non-Latin scripts, and LaTeX math.
via MarkTechPost