Decoding Documents: The Power of Agentic Document Extraction Gen2
The central development is this: In today’s data-driven world, extracting precise information from diverse documents remains a significant challenge. Unstructured data often hides critical insights, making efficient and accurate processing essential for businesses.
Table of Contents
- Decoding Documents: The Power of Agentic Document Extraction Gen2
- Introducing the DPT-3 Model Family: Pro and Verity
- A Revolutionary Approach to Pricing
- Structured Outputs: Beyond Simple Chunks
- Atomic Grounding: Unlocking Unprecedented Precision
- Deployment and Availability
- Expert Perspective
- Frequently Asked Questions
- Key Takeaways
- DPT-3 Verity: Precision for High-Volume Digital Documents
- DPT-3 Pro: Advanced Understanding for Complex Layouts
- Why is Agentic Document Extraction Gen2 important?
- What impact could Agentic Document Extraction Gen2 have?
- What should readers watch next with Agentic Document Extraction Gen2?
- How does this relate to gen2?
LandingAI has stepped up to this challenge with the release of Agentic Document Extraction (ADE) Gen2, a complete overhaul of its document intelligence platform. This new generation promises enhanced affordability, outputs tailored for AI agents, and unprecedented atomic grounding.
Meanwhile, ADE Gen2 represents a fundamental shift in how documents are processed. While its predecessor, Gen1, viewed documents as flat lists of data chunks, Gen2 intelligently treats them as a hierarchical tree structure. This foundational change allows for more nuanced understanding, more precise data extraction, and a more granular pricing model based on characters returned rather than just pages.
Introducing the DPT-3 Model Family: Pro and Verity
At the heart of ADE Gen2 are two new, specialized parsing models: DPT-3 Pro and DPT-3 Verity. This dual-model approach allows organizations to select the right tool for their specific workload, optimizing both performance and cost.
DPT-3 Verity: Precision for High-Volume Digital Documents
In practical terms, DPT-3 Verity is engineered for digitally created documents, offering deterministic transcription. It excels with high-volume text, tables, and simple form fields, providing a bounding box and a confidence score for every word. This model is designed for efficiency and cost-effectiveness, charging significantly fewer credits than its Pro counterpart.
DPT-3 Pro: Advanced Understanding for Complex Layouts
For more complex documents, DPT-3 Pro takes the lead. This model first analyzes the page layout before transcribing words, intelligently detecting various block types—from tables and figures to marginalia and signatures.
It handles scanned pages, handwriting, non-Latin scripts, and even LaTeX math, returning content in logical reading order. LandingAI plans to introduce automated routing between DPT-3 Verity and Pro in the fall of 2026, further streamlining the selection process.
A Revolutionary Approach to Pricing
For example, One of the most impactful changes in ADE Gen2 is its redesigned pricing structure. Moving away from a flat per-page fee, the new model combines a page component with an output character component, offering a more precise reflection of usage.
- DPT-3 Pro (Priority Tier): Bills 1 credit per page plus 0.5 credits per 1,000 output characters.
- DPT-3 Verity (Priority Tier): Bills 0.3 credits per page plus 0.2 credits per 1,000 output characters.
The standard tier halves these rates, catering to pipelines that can tolerate asynchronous processing. This granular billing model, combined with service tiers, is projected to deliver significant cost reductions, potentially parsing documents for less than one cent per page with Verity on the standard tier for suitable workloads. However, businesses are encouraged to benchmark their own document mixes, as dense, character-rich pages might still incur higher costs than before.
Structured Outputs: Beyond Simple Chunks
That said, ADE Gen2’s Parse v2 response is meticulously structured with three top-level fields: markdown in reading order, metadata, and structure. The core of this structure is a document node, with children representing pages, and pages containing various blocks. This hierarchical approach offers a much richer understanding of document content.
Block types are extensive, including text, table, table_cell, figure, marginalia, attestation, logo, card, and scan_code. Each block comes with a semantic ID and a grounding object that specifies its page number, range within the markdown string, and a normalized bounding box. Furthermore, markdown output is standardized; for instance, figures use specific HTML elements with generated prose isolated within <description> tags, and tables are shipped as HTML by default to preserve complex merged cells.
Atomic Grounding: Unlocking Unprecedented Precision
Interestingly, Perhaps the most transformative capability of ADE Gen2 is its atomic grounding. Every leaf block now carries an atomic_grounding array.
For DPT-3 Pro, this means grounding to each visual line, while DPT-3 Verity grounds to every single word, complete with a confidence value (from 0 to 1). This confidence score, derived from the lowest per-character score, provides a crucial signal for routing uncertain transcriptions for human review.
This level of precision has profound implications. Extracted fields can now be traced back to a specific word on a specific page, making tasks like PII redaction by coordinate, robust document diffing, and the creation of highly accurate reviewer user interfaces not just possible, but buildable with confidence.
Deployment and Availability
However, ADE Gen2 is generally available now, with developers able to explore its capabilities in the free ADE playground. Enterprise clients have flexible deployment options, including US or EU cloud environments, their own Virtual Private Clouds (VPC) on AWS, Azure, or Google Cloud, integration within Snowflake, or even on-premises, including air-gapped environments.
It’s important for existing users to note that Gen1 client code will require migration to work with Gen2 endpoints, reflecting the significant architectural advancements in this new release.
Expert Perspective
A practical read on Agentic Document Extraction Gen2 starts with gen2. That is where the earliest effects are likely to show up if this development keeps building.
What happens next will come down to adoption speed, policy response, and execution quality. That combination could make Agentic Document Extraction Gen2 a meaningful reference point across page.
For decision-makers, the useful lens is not the headline alone but how documents changes priorities once organizations have to respond.
Frequently Asked Questions
Why is Agentic Document Extraction Gen2 important?
Decoding Documents: The Power of Agentic Document Extraction Gen2The central development is this: In today’s data-driven world, extracting precise information from diverse documents remains a significant challenge.
What impact could Agentic Document Extraction Gen2 have?
Unstructured data often hides critical insights, making efficient and accurate processing essential for businesses.LandingAI has stepped up to this challenge with the release of Agentic Document Extraction (ADE) Gen2, a complete overhaul of its document intelligence platform.
What should readers watch next with Agentic Document Extraction Gen2?
This new generation promises enhanced affordability, outputs tailored for AI agents, and unprecedented atomic grounding.Meanwhile, ADE Gen2 represents a fundamental shift in how documents are processed.
How does this relate to gen2?
It connects because the article frames gen2 as one of the clearest areas where the topic may be felt in practice.
Key Takeaways
- DPT-3 Pro and DPT-3 Verity replace DPT-2 as the primary parsing models, offering specialized capabilities.
- Billing has shifted from a flat per-page rate to a more granular page plus output character rate.
- DPT-3 Pro grounds to the line level, while DPT-3 Verity provides word-level grounding with confidence scores.
- The response structure is now a robust document, page, and block tree, replacing the simpler ‘chunks’.
- Migration is required for existing Gen1 client code to integrate with Gen2 endpoints.



























