LlamaIndex launched OpenDocRouter, a unified API for document parsing. It supports PDFs, PNGs, JPEGs, or URLs as input. The API offers synchronous and asynchronous processing modes. Synchronous mode is limited to 50 pages. All models can return grounded bounding boxes and layout elements when requested. OpenDocRouter uses token-based billing with model-specific rates. It includes frontier and open-source models at launch. New models will be added and benchmarked via ParseBench. Layout mode adds a $0.20 per million tokens fee. Failed pages are not charged. OpenDocRouter is live now.
What changed
Users had to manage prompts, rate limits, deployments, and benchmarking for each OCR model separately.
What it unlocks
Switching between document parsing models via a single API call.
- $0.10 to $4.00 per 1M input tokens
- $0.01 to $20.00 per 1M output tokens
- 500 pages or 50MB max per request
- 50 pages max for sync responses
What you need to act on it
- an API key
- a top-up of at least $25
Sources