Polish OCR Benchmarks – Results
Collection
Collection of OCR benchmark result datasets focused on Polish-language and difficult real-world documents. • 6 items • Updated • 2
model string | elo int64 | params string | wins int64 | losses int64 | ties int64 | win_pct int64 | elo_low int64 | elo_high int64 |
|---|---|---|---|---|---|---|---|---|
clearocr.com/clearocr-api | 1,793 | 379 | 56 | 1 | 87 | 1,754 | 1,840 | |
deepseek-ai/DeepSeek-OCR | 1,617 | 4B | 291 | 144 | 1 | 67 | 1,585 | 1,653 |
lightonai/LightOnOCR-2-1B | 1,471 | 1B | 203 | 232 | 0 | 47 | 1,439 | 1,502 |
baidu/Qianfan-OCR | 1,311 | 4.7B | 108 | 327 | 0 | 25 | 1,276 | 1,344 |
rednote-hilab/dots.ocr | 1,309 | 1.7B | 107 | 329 | 0 | 25 | 1,272 | 1,344 |
OCR Bench Results: Polish insurance terms and OWU benchmark
VLM-as-judge pairwise evaluation of OCR models on Polish insurance / OWU-style documents. Results depend strongly on document type, so this should be read as a document-specific benchmark rather than a universal OCR ranking.
This benchmark focuses on dense Polish insurance and legal text, including OWU-style documents with small fonts, long paragraphs, numbered sections and structured lists.
Leaderboard
| Rank | Model | Params | ELO | 95% CI | Wins | Losses | Ties | Win% |
|---|---|---|---|---|---|---|---|---|
| 1 | clearocr.com/clearocr-api | 1793 | 1754–1840 | 379 | 56 | 1 | 87% | |
| 2 | deepseek-ai/DeepSeek-OCR | 4B | 1617 | 1585–1653 | 291 | 144 | 1 | 67% |
| 3 | lightonai/LightOnOCR-2-1B | 1B | 1471 | 1439–1502 | 203 | 232 | 0 | 47% |
| 4 | baidu/Qianfan-OCR | 4.7B | 1311 | 1276–1344 | 108 | 327 | 0 | 25% |
| 5 | rednote-hilab/dots.ocr | 1.7B | 1309 | 1272–1344 | 107 | 329 | 0 | 25% |
Details
- Task: OCR (Optical Character Recognition)
- Language: Polish
- Document type: Polish insurance terms and OWU-style legal text
- Original dataset:
byczong/pl-insurance-terms-struct - Source dataset:
Lukaszl/clearocr-benchmark-pl-insurance-terms-v1 - Judge: Qwen3.5-35B-A3B
- Comparisons: 1089
- Method: Bradley-Terry MLE with bootstrap 95% CIs
About clearOCR
clearOCR is an OCR API for extracting text from PDFs, scans and document images, with a strong focus on Polish and English documents.
New accounts currently receive:
- 1,000 free single-image OCR runs
- valid for 30 days
API access is available via the clearOCR website:
Configs
load_dataset("Lukaszl/clearocr-benchmark-pl-insurance-terms-v1-results")— leaderboard tableload_dataset("Lukaszl/clearocr-benchmark-pl-insurance-terms-v1-results", name="comparisons")— full pairwise comparison logload_dataset("Lukaszl/clearocr-benchmark-pl-insurance-terms-v1-results", name="metadata")— evaluation run history
Generated by ocr-bench
- Downloads last month
- 95