How to Choose a Container Number OCR API in 2026

Why Demos Mislead

Most container number OCR demos use clean, flat images. Production traffic includes glare, blur, cropped edges, and unusual issuers, and those images decide your real success rate and cost.

Evaluate on Your Own Images

Collect a consented sample of real customer images, run each candidate API on the same set, and compare field-level accuracy and failure behaviour. StructOCR offers 200 free credits for this; see the API docs and the free scanner.

Where This Fits

• Accuracy: Measure per-field accuracy, not just “document read”, and check how often the API returns plausible but wrong values.
• Failure handling: Prefer APIs that return clear errors you can turn into retake prompts over ones that return partial guesses.
• Cost per success: Compare cost per successful result, including whether failed requests are billed.

Benchmark Script Starting Point

Call the API for each test image and store the JSON for side-by-side comparison.

Prerequisite: Python 3.9+, `pip install requests`, and a STRUCTOCR_API_KEY

import base64
import os
import requests

with open("container.jpg", "rb") as image_file:
    image = base64.b64encode(image_file.read()).decode("ascii")

response = requests.post(
    "https://api.structocr.com/v1/container",
    headers={"x-api-key": os.environ["STRUCTOCR_API_KEY"]},
    json={"img": image},
    timeout=30,
)
response.raise_for_status()
data = response.json()["data"]
print("container_number:", data.get("container_number"))
print("is_valid:", data.get("is_valid"))
print("confidence:", data.get("confidence"))

What a Structured Response Looks Like

Illustrative field names from the StructOCR response for this document type.

{
  "success": true,
  "data": {
    "container_number": "…",
    "is_valid": "…",
    "confidence": "…"
  }
}