OpenParser
Cheapest OCR API for document layout & tables
OpenParser is an OCR API specializing in layout and table parsing, powered by the open-weight model PaddleOCR-VL-1.6. It offers the lowest public list rate for structure-preserving OCR at $1 per 1,000 pages after 1,000 free pages, significantly undercutting competitors like AWS Textract and Azure Layout. Designed for developers who need reliable extraction of blocks, tables, and markdown from documents such as invoices, statements, and forms, OpenParser provides a transparent, page-based pricing model with no minimums.
The service solves the problem of high-cost, complex OCR for layout-heavy documents. By separating parse and extract metering—pages for structure recognition and tokens for schema filling—it gives users control over spending. The same API supports sync, async, and batch requests, and includes a playground for testing documents before integration. OpenParser is built for teams that require confidence scores, block citations, and the ability to self-host the underlying model.
Key Features
Flat Page Pricing
Pay $1 per 1,000 pages for layout-aware parsing, with the first 1,000 pages free. No minimums, and the rate applies to sync, async, and batch requests.
Open-Weight Model (PaddleOCR-VL-1.6)
Uses the PaddleOCR-VL-1.6 model, which tops independent document parsing benchmarks and is open-weight, allowing self-hosting as an alternative to the API.
Block-Level OCR with Confidence Scores
Retrieve typed blocks (text, table, etc.) with bounding boxes and confidence scores, enabling you to gate uncertain pages before downstream processing.
Separate Parse and Extract Meters
Parsing is billed per page for blocks and markdown. Extraction adds token costs based on the LLM model you choose, keeping costs transparent.
Block Citations for Extracted Values
Extracted schema values cite the original block IDs from the parse, making audit and review easy and preventing silent errors from mismatched geometry.
Playground and Console
Test documents live in the Playground without code, inspect blocks and markdown, and manage API keys, billing, and team roles from the console.
Same API for Parse and Extract
Use a single API endpoint (/parse and /extract) with consistent metering. Supports sync, async, and batch modes with idempotency keys.
Use Cases
- Extract tables from invoices and purchase orders
- Parse structured fields from bank statements and payment authorizations
- Process multi-page forms with checkboxes, signatures, and dates
- Automate data entry from receipts and expense reports
- Build document processing pipelines for accounting and finance teams
- Validate OCR confidence on uncertain documents before billing
- Create searchable markdown from scanned PDF reports
- Self-host the model for high-volume or data-sensitive environments
FAQ
What is the pricing for OpenParser?
Parsing costs $1 per 1,000 pages after a one-time free allotment of 1,000 pages. Extraction adds per-token costs based on the LLM model you select. There are no minimums.
What model does OpenParser use?
It uses PaddleOCR-VL-1.6, an open-weight model that ranks highest on independent document parsing benchmarks.
Can I self-host the OCR model instead of using the API?
Yes, PaddleOCR-VL is open-weight and can be self-hosted, replacing page fees with your own GPU and operational costs.
How does extraction differ from parsing?
Parsing returns blocks and markdown from the document. Extraction fills a user-defined schema by applying an LLM to those blocks, and is billed per token.
Does OpenParser support asynchronous or batch processing?
Yes, the API supports sync, async, and batch requests for the /parse and /extract endpoints.
How can I verify the accuracy of OCR results?
Each parsed block includes a confidence score and bounding box. You can route low-confidence pages for manual review before they are used downstream.