OpenParser

About OpenParser

OpenParser is a document-parsing API whose selling point is model choice rather than a single house OCR engine. Nine hosted OCR models sit behind one request shape: PaddleOCR-VL, two generations of Mistral OCR, Azure Document Intelligence Layout and Read, Google enterprise document engines and AWS Textract among them. You change the `ocr_model` field and nothing else about your integration, which is a direct answer to the usual failure mode of document pipelines — the engine you standardised on turns out to be wrong for one document class and switching means rewriting the integration. Two endpoints carry the product. POST /parse returns a typed `openparser@1` block structure plus markdown, where blocks carry a kind (text, table, field, checkbox, signature), page number, bounding box and confidence. POST /extract fills a schema you supply and cites the block ids each value came from, so a downstream reviewer can trace any extracted field back to its position on the page. Sync, async and batch requests share the same request shape, with GET /jobs/{id} for async status and GET /files/{id}/content for reusable file bytes. Idempotency keys are supported on parse. Pricing is per page and published per model, which makes cost routing an explicit design decision: send bulk general documents to the cheapest open-weight engine and reserve the expensive enterprise engines for the layout-heavy exceptions.

Key Features

OpenParser Pros & Cons

Tags