Project AURIONAI Use-Cases Research & Innovation

Real5-OmniDocBench

Document parsing on real-world captures

multimodalactive
9scores
9models
0independent
0maintainer-run
9vendor-reported

Definition

Version
not in the notes
Category
multimodal
Maintainer
not in the notes
Status
active
Home page
not in the notes

Sources

  1. arxiv.org/pdf/2606.03264 https://arxiv.org/pdf/2606.03264

Scores at a Glance

Score, PaddleOCR-VL-1.6 paper table

Higher is better for this unit.

1BaiduPaddleOCR-VL-1.6Baiduoverall
93.19 score
vendor-reported2BaiduPaddleOCR-VL-1.5Baiduoverall
92.05 score
vendor-reported3Z.ai (Zhipu)GLM-OCRZ.ai (Zhipu)overall
90.32 score
vendor-reported4Google DeepMindGemini 3 ProGoogle DeepMind, Googleoverall
89.24 score
vendor-reported5OpenDataLabMinerU2.5-ProOpenDataLaboverall
88.96 score
vendor-reported6rednote (Xiaohongshu)dots.ocrrednote (Xiaohongshu)overall
86.38 score
vendor-reported7OpenDataLabMinerU2.5 (MinerU-2.5)OpenDataLaboverall
85.61 score
vendor-reported8DeepSeekDeepSeek-OCRDeepSeekoverall
73.99 score
vendor-reported9DeepSeekDeepSeek-OCR 2DeepSeekoverall
73.01 score
vendor-reported
0255075100
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.

Scores

9 of 9 scores shown real5-omnidocbench-scores.csv
ModelScoreWho ran itSettingView or populationDateSource
BaiduPaddleOCR-VL-1.693.19 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
BaiduPaddleOCR-VL-1.592.05 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
Z.ai (Zhipu)GLM-OCR90.32 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
Google DeepMindGemini 3 Pro89.24 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
OpenDataLabMinerU2.5-Pro88.96 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
rednote (Xiaohongshu)dots.ocr86.38 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
OpenDataLabMinerU2.5 (MinerU-2.5)85.61 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
DeepSeekDeepSeek-OCR73.99 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
DeepSeekDeepSeek-OCR 273.01 scorevendor-reportedoverallPaddleOCR-VL-1.6 paper table2 Jun 2026arxiv.org
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Real5-OmniDocBench", https://aurion.semai.id/benchmarks/real5-omnidocbench/, records of October 2026.