clearOCR (TeamQuest): Independent Benchmark Results.

We ran the full OmniDocBench benchmark (1348 images) on TeamQuest's clearOCR API. Here's what we found.

Independently verified.

CodeSOTA ran the full OmniDocBench evaluation suite on December 19, 2025. We processed 1,348 images through TeamQuest's clearOCR API and computed metrics using the official evaluation tools.

Traditional OCR, not document understanding.

clearOCR is a traditional OCR solution that extracts plain text from images. It does not recognize tables as structured data or mathematical formulas as LaTeX. This significantly impacts OmniDocBench composite scores which weight tables and formulas.

The four metrics that matter.

Verified scores against the field.

OmniDocBench is a comprehensive document parsing benchmark with 1,355 pages across 9 document types. The composite score formula: ((1-TextEditDist)*100 + TableTEDS + FormulaCDM) / 3

Lower is better for Edit Distance metrics. CodeSOTA verification date: December 19, 2025.

Strong on academic, weak on newspapers.

clearOCR performs best on research reports and academic papers, struggles significantly with newspapers:

English leads, Chinese trails.

Polish OCR from TeamQuest.

clearOCR is a Polish OCR service developed by TeamQuest Sp. z o.o. It's designed for extracting text from documents, with a focus on Polish and English language support.

Calling the API.

import requests
API_URL = "https://clearocr.teamquest.pl:60213/extract-document-parser"
headers = {
    "CLEAR-OCR-API-USER": "your-username",
    "CLEAR-OCR-API-KEY": "your-api-key",
    "CLEAR-OCR-API-VERSION": "0.1"
}
with open("document.png", "rb") as f:
    files = {"file": f}
    response = requests.post(
        API_URL,
        headers=headers,
        files=files,
        verify=False  # Self-signed cert
    )
result = response.json()
print(result.get("result", {}).get("text", ""))

Plain text, no markdown structure.

clearOCR outputs plain text without markdown structure:

Who Am I?
- Min-Te Sun (Peter) Sun
  - An associate professor of Computer Science & Information Engineering
  - National Central University
  - Studied in US for a long time (from 1993 ~ 2002)
  - Worked as a CS professor at Auburn University, Alabama
Is Good English Important?
- For Individual
  - Reading - lifelong learning to improve your competitiveness
  - Writing - emails with foreign friends/co-workers/customers
  - Listening - Q/A with foreign customers/boss/etc

Fit for purpose.

Composite Score: 31.7.

Composite Score: 31.7 — Low on OmniDocBench due to missing table/formula support.

clearOCR is a traditional OCR solution that extracts text only. It performs reasonably well for pure text extraction (84.6% overall, up to 95% on clean documents), but lacks modern document understanding capabilities.

The low composite score (31.7 vs Mistral's 79.75) is primarily because OmniDocBench weights tables and formulas equally with text. For pure text extraction tasks, clearOCR is a viable option, especially for Polish documents.

Best use case: Simple text extraction from clean documents where you don't need table structure or LaTeX formulas.

Continue reading.