Oriel — AI & technology worth noticing, every day.

Xiaomi-OCR-0 Technical Report

Paper · https://arxiv.org/abs/2609.36136

Compact OCR-specific vision-language models achieve strong document parsing performance, but often rely on costly supervision and focus primarily on visual-text reconstruction. We introduce Xiaomi-OCR-0, a unified 0.8B model for document parsing and OCR-centric understanding. We build an approximately 170M-sample OCR-centric corpus using an automated data engine that combines expert consensus, render-based verification, and targeted synthesis. Starting from Qwen3.5-0.8B, our progressive training recipe combines Q-Mask-based text anchoring, continued pretraining, and mixed-task reinforcement learning (Mix-RL). Xiaomi-OCR-0 achieves 95.24 on Real5-OmniDocBench, 96.83 on OmniDocBench v1.6, and 87.94 on Wild-OmniDocBench, while reaching an average score of 83.2 across five OCR-oriented VQA benchmarks. Ablations further show that, with sufficient parsing training, OCR-centric understanding supervision provides additional gains for document parsing. Homepage: https://huggingface.co/spaces/SeerRay-Lab/Xiaomi-OCR-0.

First seen by Oriel on .

AI summary

Xiaomi-OCR-0 is a unified 0.8B model for document parsing and OCR-centric understanding, trained on a large automated corpus and reporting strong results across document and OCR-oriented benchmarks.

  • Its approximately 170M-sample corpus combines expert consensus, render-based verification, and targeted synthesis.
  • Training starts from Qwen3.5-0.8B and combines Q-Mask-based text anchoring, continued pretraining, and mixed-task reinforcement learning (Mix-RL).
  • The model scores 95.24 on Real5-OmniDocBench, 96.83 on OmniDocBench v1.6, and 87.94 on Wild-OmniDocBench.
  • It averages 83.2 across five OCR-oriented VQA benchmarks; ablations find OCR-centric understanding supervision further improves document parsing when parsing training is sufficient.

#ocr #vision-language models #document parsing

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Points

Score reported by the source.

  • Hugging Face Papers: 14, latest observation ; +7 over observed period since

Comments

Comment count reported by the source.

  • Hugging Face Papers: 0, latest observation ; 0 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Papers2609.36136https://arxiv.org/abs/2609.36136 Read discussion