AI OCR API in India

Optical character recognition (OCR) with AI

Modern OCR goes beyond drawing boxes around text. unoblox lets you build document-parsing systems on vision-capable models — extracting not just raw text but structured fields — through one OpenAI-compatible endpoint (https://api.unoblox.ai/v1), billed in rupees on a GST invoice.

Why AI-powered OCR beats rule-based engines

Traditional OCR stumbles on handwriting, skewed scans, and mixed-language documents. Vision-capable LLMs read layout and context together — they can tell a total is a total because it sits next to "Grand Total", not just because it's a number in a box.

Recommended models

Model	Input ₹/1M	Output ₹/1M	Best for
GPT-4o	₹252	₹1008	Fastest, strong general accuracy
Claude Opus	₹504	₹2520	Dense, multi-page documents
Qwen3 Max	₹120.95	₹604.77	Cost-effective, solid accuracy

Prices per 1M tokens — live rates at /models.

Extract raw text from an image

from openai import OpenAI
import base64

client = OpenAI(
    api_key="ub-gw-...",
    base_url="https://api.unoblox.ai/v1"
)

with open("document.png", "rb") as f:
    image_data = base64.b64encode(f.read()).decode()

response = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[{
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{image_data}"}},
            {"type": "text", "text": "Extract all text from this image."}
        ]
    }]
)

print(response.choices[0].message.content)

Extracting structured fields, not just text

For invoices, receipts, and forms you usually want fields, not a text blob. Add a schema to the prompt and turn on JSON mode:

response = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[{
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{image_data}"}},
            {"type": "text", "text": "Extract as JSON: {vendor, invoice_number, date, line_items: [{description, qty, amount}], total}"}
        ]
    }],
    response_format={"type": "json_object"}
)

What it costs

A single invoice image typically resolves to 500–1,500 input tokens (image + prompt) and 100–300 output tokens for the extracted fields. At GPT-4o rates, that's roughly ₹0.15–₹0.50 per document — processing 10,000 invoices a month lands somewhere between ₹1,500 and ₹5,000 depending on document complexity.

Use cases

• Invoice digitization and accounts-payable data entry
• Receipt scanning for expense management
• Handwritten form processing (surveys, applications)
• Multilingual document extraction and translation
• ID and passport field verification for onboarding

Frequently asked questions

Q: Can unoblox handle handwritten text? A: Yes, GPT-4o and Claude Opus both read handwriting reasonably well, though results depend heavily on image quality — scan at a higher resolution if extraction looks unreliable.

Q: Does my document data leave unoblox's platform? A: Requests are billed and logged through unoblox's India-based gateway; the model you choose (GPT-4o, Claude Opus, Qwen3 Max) determines where the actual image processing happens — see each model's page on /models for details before sending sensitive documents.

Q: How much does OCR cost per document? A: A typical single-page document uses 500–1,500 tokens. At GPT-4o rates that's roughly ₹0.15–₹0.50 per page — see the cost breakdown above for volume estimates.

Q: Can I process multiple images in one request? A: Yes, add multiple image blocks to the same content array, or loop over documents. There's no dedicated batch endpoint yet, but raw throughput is high.

Q: Which models support vision? A: GPT-4o, Claude Opus, Claude Sonnet, and Qwen3 Max all accept images. Check each model's card on /models to confirm current vision support before building against it.

Q: Can I get structured JSON instead of a text blob? A: Yes — set response_format to json_object and describe the schema you want in your prompt, as shown above.

Get started in rupees → https://unoblox.ai/sign-in