LandingAI Releases Agentic Document Extraction Gen2

LandingAI has launched Agentic Document Extraction (ADE) Gen2, a significant overhaul of its document intelligence platform built around a new model family named DPT-3. This updated version moves beyond Gen1's approach of treating documents as flat lists of chunks, instead processing them as a tree structure. ADE Gen2 prices its services based on the number of characters returned, not by the page, and crucially, grounds every answer to a specific line or word within the original document. The release is framed by LandingAI around three core themes: enhanced affordability, agent-ready outputs, and atomic grounding.

ADE Gen2 is now generally available, with developers able to start using it for free in the ADE playground. For enterprise clients, the platform can be deployed in various cloud environments, including AWS, Azure, and Google Cloud within their own Virtual Private Clouds (VPCs), directly inside Snowflake, or on-premises, even in air-gapped configurations. A key innovation in Gen2 is the split of the parsing process into two distinct models, DPT-3 Verity and DPT-3 Pro, allowing for more efficient workload management and cost optimization. DPT-3 Verity is designed for digitally created documents, providing deterministic transcription with bounding boxes and confidence scores for every word, making it suitable for high-volume text, tables, and simple form fields. In contrast, DPT-3 Pro analyzes page layout before text, capable of detecting block types such as tables, figures, marginalia, and signatures, returning them in reading order. It also handles more complex inputs like scanned pages, handwriting, non-Latin scripts, and LaTeX mathematical equations.

LandingAI has stated that DPT-3 Verity incurs approximately 40% of the credit cost compared to DPT-3 Pro. The company also plans to introduce automated routing between these two models in the fall of 2026. The pricing structure represents a substantial shift from DPT-2, where each page incurred a flat rate of 3 credits. Under the DPT-3 model, credit consumption is calculated as the sum of a page component and an output character component. Specifically, on the priority tier, DPT-3 Pro charges 1 credit per page plus 0.5 credits per 1,000 output characters. DPT-3 Verity is priced at 0.3 credits per page plus 0.2 credits per 1,000 output characters. The standard tier offers a 50% reduction on both rates. For example, a 12-page Pro parse that returns 48,120 characters would cost 36.1 credits on the priority tier, with the standard tier costing roughly half that amount. All totals are rounded up to the nearest 0.1 credit.