API Services for PDF Extraction

API Services for PDF Extraction

Mistral OCR API (Recommended)

Best for: Complex layouts, scanned documents, multilingual content, math formulas.

Pricing: ~$1 per 1000 pages (very cost-effective)

Accuracy: 95%+ overall, 98%+ on scanned documents

Features

• Outputs clean markdown preserving document structure
• Table recognition (96% accuracy)
• Math equation support (94% accuracy)
• Multilingual support (89% accuracy)
• Processes up to 2000 pages/minute

API Usage

from mistralai import Mistral
import base64

client = Mistral(api_key="your-key")

# From URL
response = client.ocr.process(
    model="mistral-ocr-latest",
    document={
        "type": "document_url",
        "document_url": "https://example.com/doc.pdf",
    }
)

# From local file (base64)
with open("doc.pdf", "rb") as f:
    content = base64.standard_b64encode(f.read()).decode()

response = client.ocr.process(
    model="mistral-ocr-latest",
    document={
        "type": "base64",
        "base64": content,
    }
)

# Access results
for page in response.pages:
    print(page.markdown)

Endpoint

• SDK: client.ocr.process
• REST: POST https://api.mistral.ai/v1/ocr

OpenAI Vision (GPT-4V)

Best for: When you already have OpenAI API access and need to extract from image-heavy PDFs.

Approach: Convert PDF pages to images, send to GPT-4V for extraction.

Pricing: ~$0.01-0.03 per page (depends on image size and detail level)

Usage Pattern

from openai import OpenAI
from pdf2image import convert_from_path
import base64
from io import BytesIO

client = OpenAI()

def pdf_page_to_base64(image):
    buffer = BytesIO()
    image.save(buffer, format="PNG")
    return base64.b64encode(buffer.getvalue()).decode()

# Convert PDF to images
images = convert_from_path("doc.pdf", dpi=150)

for i, img in enumerate(images):
    b64 = pdf_page_to_base64(img)
    
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "user",
            "content": [
                {"type": "text", "text": "Extract all text from this document page. Preserve structure and formatting as markdown."},
                {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{b64}"}}
            ]
        }]
    )
    print(response.choices[0].message.content)

Limitations

• Requires converting PDF to images first
• Higher cost per page than Mistral OCR
• May struggle with dense text

Google Cloud Document AI

Best for: Enterprise workflows, high-volume processing, Google Cloud integration.

Pricing: $1.50 per 1000 pages (Form Parser), varies by processor type

Processors

• Document OCR: General text extraction
• Form Parser: Structured form data
• Invoice Parser: Specialized invoice extraction
• Custom Document Extractor: Train on your document types

Not covered in detail here - see Google Cloud documentation.

AWS Textract

Best for: AWS ecosystem integration, form extraction, table detection.

Pricing: $1.50 per 1000 pages (text), $15 per 1000 pages (tables/forms)

Features

• DetectDocumentText: Basic OCR
• AnalyzeDocument: Tables, forms, key-value pairs
• AnalyzeExpense: Receipt/invoice parsing

Not covered in detail here - see AWS Textract documentation.

Azure Document Intelligence

Best for: Enterprise workflows, highest accuracy on structured documents.

Formerly "Azure Form Recognizer". Ranked #1 in several 2025 OCR benchmarks.

Pricing: ~$1.50 per 1000 pages (Read model), higher for specialized models

Features

• Prebuilt models for invoices, receipts, IDs, tax forms
• Custom model training
• Excellent table and form extraction
• Strong multilingual support

Models

• Read: General OCR
• Layout: Structure + tables
• Invoice/Receipt/ID: Specialized extractors
• Custom: Train on your documents

Not covered in detail - see Azure documentation.

Comparison Summary

Recommendation: Start with Mistral OCR for best price/performance ratio. Use Azure Document Intelligence if you need highest accuracy on forms/invoices.