Document OCR

In one line

Send a scan or photo of a document and get its text back as clean Markdown, with headings, tables and line items preserved, ready to search, store or pass to another model.

The route is paranine/DeepSeek-OCR2, on the same /v1/chat/completions endpoint and the same key as every other route.

Example: digitise a supplier invoice

Input: a scanned or photographed invoice (PNG, JPEG or WebP).

Output (an example of the shape, not a real capture):

# INVOICE

**Invoice No:** INV-2026-0412  **Date:** 7 October 2026
**Bill to:** Example Traders (Pvt) Ltd, Lahore

| Item                     | Qty | Unit price (Rs) | Amount (Rs) |
|--------------------------|----:|----------------:|------------:|
| Office chairs            |  12 |           8,500 |     102,000 |
| Standing desks           |   4 |          42,000 |     168,000 |
| **Subtotal**             |     |                 | **270,000** |
| Sales tax (18%)          |     |                 |      48,600 |
| **Total**                |     |                 | **318,600** |

What it costs: a request holds at most 8,192 tokens, so a page costs at most about 7 paisa (Rs 74 per 1,000 pages) at the prices below, and usually less.

Pricing

You pay for the tokens the model actually processes: the page image counts as input, the extracted text as output. Each response reports both in usage.

Code

Put your API key in P9_API_KEY. Each example reads invoice.png from the current folder and prints the Markdown.

export P9_API_KEY="sk_p9_…"   # your API key

curl -sS https://sovereign.paranine.com/v1/chat/completions \
  -H "Authorization: Bearer $P9_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @- <<EOF
{
  "model": "paranine/DeepSeek-OCR2",
  "temperature": 0,
  "max_tokens": 4096,
  "messages": [{
    "role": "user",
    "content": [
      {"type": "image_url", "image_url": {"url": "data:image/png;base64,$(base64 < invoice.png | tr -d '\n')"}},
      {"type": "text", "text": "Convert the document to markdown."}
    ]
  }]
}
EOF

Good to know

• One page per request. A request holds up to 8,192 tokens, input and output together. Keep max_tokens at 6,000 or below for one page.
• Formats: PNG, JPEG or WebP. For a JPEG, use data:image/jpeg;base64, in the image URL.
• Size: up to about 1.4 MB per image (requests are limited to 2 MB, and base64 adds a third).
• Not in the playground. The dashboard playground sends text only; call this route from code.