Run PaddleOCR-VL
without your own GPUs.
OpenParser hosts PaddleOCR-VL-1.6 behind a REST API. You send PDFs and images. You get typed layout blocks or markdown, then schema extraction with block citations.
Hosted PaddleOCR-VL behind a REST API.
Parse with PaddleOCR-VL-1.6, extract against your schema, then poll jobs and retrieve files, with the same contract as the main API.
/parserequestcurl https://api.openparser.dev/parse \
-H "Authorization: Bearer op_live_…" \
-H "Idempotency-Key: $(uuidgen)" \
-F 'request={"ocr_model":"paddleocr-vl-1.6","output_format":"openparser@1"};type=application/json' \
-F "file=@statement.pdf"{
"output_format": "openparser@1",
"document_id": "doc_3f9a2c",
"page_count": 1,
"markdown": "## Payment Authorization\n| Item | Amount |\n| …",
"blocks": [
{ "index": 0, "page_number": 1, "kind": "text",
"text": "## Payment Authorization" },
{ "index": 1, "page_number": 1, "kind": "table",
"table_html": "<table>…</table>",
"bbox": { "left": 14, "top": 53, "right": 146, "bottom": 123 },
"confidence": 0.98 }
],
"regions": [],
"contents": [],
"chunks": []
}Try PaddleOCR-VL in the console.
Upload a document in the playground, create environment keys, and review page usage. OpenParser hosts the model; you call the API.
{ total: "$4,318.20" } b4What you get beyond raw OCR.
PaddleOCR-VL handles layout. OpenParser wraps it in a stable contract, job modes, and grounded extraction.
Typed blocks and markdown
Each page returns layout blocks with kinds like text, headings, tables, and figures. Request markdown, blocks, or both in one call.
Per-block confidence
Parsed blocks include OCR confidence scores. Route low-confidence pages to review before you commit extracted values.
Pinned model version
OpenParser runs PaddleOCR-VL-1.6. You pass ocr_model in the request body. Upgrades roll out on the hosted side.
Schema extraction with citations
Point /extract at a parse job and a JSON schema. Grounding citations link each field to block_index values in the parsed document.
openparser@1 contract
Responses use one normalized output format. Block types, bounding boxes, and markdown fields stay consistent across calls.
Sync, async, and batch
Parse one file synchronously, queue a large PDF for async processing, or submit a batch of documents in one request.
Pay for the pages you parse.
The first 1,000 pages are free. After that, parsing is $1 per 1,000 pages. Extraction adds the token cost of whichever model you pick.
- No minimums
- No per-seat fees
- Sync, async, and batch
- Block citations
- Console and API keys
PaddleOCR API, answered.
Hosting, output shape, extraction, and pricing for PaddleOCR-VL-1.6 on OpenParser.
What is PaddleOCR-VL-1.6?
PaddleOCR-VL-1.6 is an open-weights vision-language model for document layout parsing. OpenParser hosts it behind a REST API so you parse PDFs and images without running your own inference stack.
What does the /parse endpoint return?
A ParsedDocument with typed layout blocks and optional markdown. Blocks include kind, text or table HTML, bounding boxes, and per-block OCR confidence.
How does schema extraction work?
Call /extract with a parse job id and a JSON schema. The response fills your fields and returns grounding metadata with citations that point to block_index values in the parsed document.
Can I process documents asynchronously?
Yes. Use /parse/async for a single file or /parse/batch for many files. Poll the job endpoint until status is complete, then fetch the result.
How much does parsing cost?
The first 1,000 parsed pages are free. After that, parsing costs $1 per 1,000 pages. Extraction adds the token cost of whichever language model you select.
Do I need my own GPUs?
No. You send files to the API with your API key. OpenParser operates the PaddleOCR-VL-1.6 inference infrastructure.
Parse your documents
with PaddleOCR-VL.
Upload a file in the playground and inspect the typed blocks, markdown, and grounded JSON. No GPU cluster required.