upstage-ocr

Upstage OCR

Extract word-level text with bounding box coordinates from images and scanned documents.

Quick Start

import os
import requests

response = requests.post(
    "https://api.upstage.ai/v1/document-digitization",
    headers={"Authorization": f"Bearer {os.environ['UPSTAGE_API_KEY']}"},
    files={"document": open("scan.pdf", "rb")},
    data={"model": "ocr"}
)
result = response.json()
print(result["pages"][0]["text"])

API Key: Always use os.environ["UPSTAGE_API_KEY"]. Get your key at console.upstage.ai.

---

Endpoints

Mode	Endpoint	Max pages	Max file size
Sync	POST /v1/document-digitization	100	50 MB
Async	POST /v1/document-digitization/async	1000	50 MB

• Request format: multipart/form-data
• Sync: returns the result in the response body (timeout 5 min).
• Async: returns a request_id; poll status and download per-batch results (batches of 10 pages).

Parameters

Parameter	Type	Required	Description
model	string	Yes	ocr (alias: ocr-250904)
document	file	Yes	Document file to process
schema	string	No	clova or google (for migration)

Limits

Item	Sync	Async
Max pages	100	1000
Max file size	50 MB	50 MB
Max pixels/page	200,000,000	200,000,000

Pick sync for ≤ 100 pages and quick (≤ 5 min) processing. Pick async for documents up to 1000 pages, when you can poll, or when the sync timeout would be hit.

Supported Formats

JPEG, PNG, BMP, PDF, TIFF, HEIC, DOCX, PPTX, XLSX, HWP, HWPX

Supported Languages

• Full support: Alphabets, Korean, Chinese characters
• Partial support: Katakana, Hiragana
• Beta: Simplified Chinese

Response Structure

{
  "api": "2.0",
  "model": "ocr-250904",
  "pages": [
    {
      "id": 0,
      "text": "Full extracted text",
      "words": [
        {
          "id": 0,
          "text": "word",
          "bounding_box": {
            "vertices": [
              {"x": 0.12, "y": 0.05},
              {"x": 0.25, "y": 0.05},
              {"x": 0.25, "y": 0.08},
              {"x": 0.12, "y": 0.08}
            ]
          },
          "confidence": 0.98
        }
      ]
    }
  ],
  "usage": {"pages": 1}
}