LandingAI has released Agentic Document Extraction, or ADE, Gen2, a rebuilt version of its document intelligence stack centered on the new DPT-3 model family. The update changes how documents are processed and billed: Gen2 treats documents as tree structures, charges based on returned characters instead of page counts, and links each answer back to a specific line or word on the page. The company said the new version is now generally available, and developers can start using it for free in the ADE sandbox. Gen2 also splits parsing into two models. DPT-3 Verity is designed for deterministic transcription of digitally created documents and returns bounding boxes and confidence scores for each word, while DPT-3 Pro reads page layouts first and identifies tables, graphics, handwriting, non-Latin characters, and LaTeX math formulas. LandingAI said Verity costs about 40% of Pro and plans to add automatic routing between the two in fall 2026. Pricing has also shifted from DPT-2’s flat 3 credits per page to a mix of page-component and output-character charges under DPT-3.