LlamaIndex launched OpenDocRouter, a hosted service that converts PDFs and images into Markdown through a shared API. Developers can change the underlying model without building separate prompts and deployment arrangements for each one. Versioned processing recipes keep model settings together, while an optional layout layer adds bounding boxes and reading order to outputs from different models.
LlamaIndex, a company developing tools for AI applications that use documents and other data, launched OpenDocRouter on October 7. The hosted service puts document-parsing models behind one application programming interface, or API. It converts PDFs and images into Markdown, a plain-text format for structured documents. Developers can switch models while keeping a common request format.[1], [2]
Each model runs a versioned recipe containing prompts, processing steps and settings. Inputs can be PDF, PNG or JPEG files, or links to those formats. Developers can receive a result directly or create an asynchronous job and check it later. Direct responses support up to 50 pages; requests are capped at 500 pages or 50 megabytes.[1]
The launch includes frontier models alongside document-oriented alternatives such as MinerU and PaddleOCR. LlamaIndex evaluates the models on its ParseBench tests for quality and processing cost.[1]
Models differ in their support for page layout. An optional grounding layer adds bounding boxes and reading order using shared classes, including titles, tables, pictures and footnotes. Billing is token-based, and the company says failed pages are not charged. LlamaIndex keeps LlamaParse as a separate managed product with enterprise controls, schema extraction and indexing.[1]
