Overview
Mistral OCR is Mistral AI's document OCR model. It returns one Markdown string per page, plus extracted images and, on newer versions, separate tables, headers and footers.
OCRRouter sends the whole file in one request and keeps the Markdown of each page. With no Mistral key set, the same model is reached through OpenParser.
Price and limits
- Model ID
- mistral/mistral-ocr
- Price
- 25 credits / page ($25 per 1,000 pages)
- Max pages per request
- 1,000
- Inputs
- PDF; PNG, JPEG and other images, sent inline as base64 data URLs
- Languages
- 100+ languages
- Positions (bbox)
- Yes, when the upstream returns them
- Up to 1,000 pages per document in OCRRouter
- Whole file is sent inline in one request
You pay only for pages that succeed. A failed attempt that falls back to another model is not charged. See pricing.
How OCRRouter calls it
- Endpoint
- POST https://api.mistral.ai/v1/ocr
- Model
- MISTRAL_OCR_MODEL, default mistral-ocr-latest
- Document
- PDF as { type: "document_url", document_url: data URL }; images as { type: "image_url", image_url: data URL }
- Auth
- Bearer token from the Mistral API key
- Fallback
- No Mistral key: OpenParser POST /parse with model OPENPARSER_MISTRAL_MODEL, default mistral-ocr-4-1
Options
What OCRRouter sets on the upstream call. You choose the model, format and pages in your request; vendor-specific options are not passed through.
- model
- From MISTRAL_OCR_MODEL, default mistral-ocr-latest
- document
- Inline data URL; no page selection
- Not set
- table_format, extract_header, extract_footer, include_image_base64, annotations: Mistral defaults apply
- Client options
- Clients cannot pass Mistral-specific options through OCRRouter
Raw response fields
Every field the upstream documents, with its type and meaning taken from the vendor's own reference. Fields marked "read by OCRRouter" are the ones our connector uses; the rest are listed so you know what the engine produces.
Mistral OCR response
Body returned by POST /v1/ocr. Field names and meanings follow Mistral's API reference and its official SDK types.
Source: Mistral API reference: OCR, retrieved 2026-11-19
pagesarray, read by OCRRouter- OCR result for each page.
pages[].indexinteger- Page index in the document, starting from 0.Example:
0 pages[].markdownstring, read by OCRRouter- Markdown text of the page. The main output.Example:
"# LEVERAGING UNLABELED DATA..." pages[].imagesarray- Images extracted from the page.
pages[].images[].idstring- Image ID, referenced from the Markdown.Example:
"img-0.jpeg" pages[].images[].top_left_xinteger | null- X of the image's top-left corner.Example:
292 pages[].images[].top_left_yinteger | null- Y of the image's top-left corner.Example:
217 pages[].images[].bottom_right_xinteger | null- X of the image's bottom-right corner.Example:
1405 pages[].images[].bottom_right_yinteger | null- Y of the image's bottom-right corner.Example:
649 pages[].images[].image_base64string | null- Base64 of the extracted image, when include_image_base64 is set.
pages[].images[].image_annotationstring | null- Image annotation as a JSON string, when requested.
pages[].tablesarray- Tables extracted from the page, when table_format is markdown or html (OCR 2512 or newer).
pages[].tables[].idstring- Table ID.
pages[].tables[].contentstring- Table content in the requested format.
pages[].tables[].formatstring- Table format, markdown or html.
pages[].hyperlinksstring[]- Hyperlinks found on the page.
pages[].headerstring | null- Page header, when extract_header is true.
pages[].footerstring | null- Page footer, when extract_footer is true.
pages[].dimensionsobject | null- Size of the page image Mistral rendered.
pages[].dimensions.dpiinteger- Dots per inch of the page image.Example:
200 pages[].dimensions.heightinteger- Page image height in pixels.Example:
2200 pages[].dimensions.widthinteger- Page image width in pixels.Example:
1700 pages[].confidence_scoresobject | null- Confidence scores, when confidence_scores_granularity is set.
pages[].blocksarray | null- Paragraph-level blocks with bounding boxes in reading order, when include_blocks is true.
modelstring- Model that produced the OCR.Example:
"mistral-ocr-2503-completion" document_annotationstring | null- Structured output in the requested format, as a JSON string.
usage_info.pages_processedinteger- Number of pages processed.Example:
29 usage_info.doc_size_bytesinteger | null- Document size in bytes.Example:
null
Shape from Mistral's API reference example; pages and text shortened.
{"pages": [{"index": 1,"markdown": "# LEVERAGING UNLABELED DATA TO PREDICT OUT-OF-D…","images": [],"dimensions": { "dpi": 200, "height": 2200, "width": 1700 }},{"index": 2,"markdown": "\nFigure 1: ...","images": [{"id": "img-0.jpeg","top_left_x": 292,"top_left_y": 217,"bottom_right_x": 1405,"bottom_right_y": 649,"image_base64": "..."}],"dimensions": { "dpi": 200, "height": 2200, "width": 1700 }}],"model": "mistral-ocr-2503-completion","usage_info": { "pages_processed": 29, "doc_size_bytes": null }}How it maps to OCRRouter output
- pages[].markdown
- pages[]: one Markdown string per page, in response order
- pages empty
- Request fails with "Provider returned no pages"
- images, tables, header, footer, dimensions
- Not used; image references stay in the Markdown as written
- Layout and confidence
- None from the direct path, so no bbox or confidence in OCRRouter output
The unified output is described in Response format.
Example request
curl https://api.ocrrouter.ai/v1/ocr \
-H "Authorization: Bearer $OCRROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "mistral/mistral-ocr",
"file_url": "https://example.com/invoice.pdf",
"format": "json"
}'Sources
Vendor documentation read on 2026-11-19.