j4xfu2mm/ocr-benchmark-results

---
license: mit
tags:
- ocr-bench
- leaderboard
source_datasets:
- j4xfu2mm/ocr-benchmark-combined
configs:
- config_name: default
data_files:
- split: train
path: data/train-*.parquet
- config_name: comparisons
data_files:
- split: train
path: comparisons/train-*.parquet
- config_name: leaderboard
data_files:
- split: train
path: leaderboard/train-*.parquet
- config_name: metadata
data_files:
- split: train
path: metadata/train-*.parquet
---

# OCR Bench Results: ocr-benchmark-combined

VLM-as-judge pairwise evaluation of OCR models. Rankings depend on document type — there is no single best OCR model.

## Leaderboard

| Rank | Model | Params | ELO | 95% CI | Wins | Losses | Ties | Win% |
|------|-------|--------|-----|--------|------|--------|------|------|
| 1 | clearocr.com/clearocr-api | | 1837 | 1801–1882 | 488 | 56 | 1 | 90% |
| 2 | deepseek-ai/DeepSeek-OCR | 4B | 1639 | 1611–1671 | 380 | 164 | 1 | 70% |
| 3 | lightonai/LightOnOCR-2-1B | 1B | 1448 | 1421–1477 | 242 | 302 | 0 | 44% |
| 4 | rednote-hilab/dots.ocr | 1.7B | 1384 | 1353–1413 | 193 | 351 | 0 | 35% |
| 5 | PaddlePaddle/PaddleOCR-VL-1.5 | | 1349 | 1321–1377 | 167 | 377 | 0 | 31% |
| 6 | baidu/Qianfan-OCR | 4.7B | 1342 | 1310–1370 | 162 | 382 | 0 | 30% |

## Details

- **Source dataset**: [`j4xfu2mm/ocr-benchmark-combined`](https://huggingface.co/datasets/j4xfu2mm/ocr-benchmark-combined)
- **Judge**: Qwen3.5-35B-A3B
- **Comparisons**: 1633
- **Method**: Bradley-Terry MLE with bootstrap 95% CIs

## Configs

- `load_dataset("j4xfu2mm/ocr-benchmark-results")` — leaderboard table
- `load_dataset("j4xfu2mm/ocr-benchmark-results", name="comparisons")` — full pairwise comparison log
- `load_dataset("j4xfu2mm/ocr-benchmark-results", name="metadata")` — evaluation run history

*Generated by [ocr-bench](https://github.com/davanstrien/ocr-bench)*