OvisOCR3 technical report released, Xiaomi open-sources similar model
The story so far
The OvisOCR3 technical report included in Hugging Face Daily Papers states that the model is a 0.8B visual language model that unifies tasks such as document parsing, general OCR, key information extraction, image-text description, and cross-page table reconstruction through multi-mentor distillation and reinforcement learning. It scored 96.74 on OmniDocBench v1.6, ranked first in OCRBench v2 English and Chinese splits, and outperformed baseline models in multiple benchmarks. These achievements are self-reported in the report. Two days later, Xiaomi announced the open-sourcing of Xiaomi-OCR-0, another 0.8B lightweight document understanding model. Weights and inference code have been made available on Hugging Face and GitHub. The single model covers three tasks: document parsing, key information extraction, and OCR visual question answering. It also provides a browser demo, MCP service, and Agent Skill toolkit for integration.
AI-written from the reports · Updated 19 days ago
Developments
2 developments
- 30 Oct, 21:15 · 1 reportsOpen-source 0.8B parameter OCR small modelITHome · AI:Xiaomi open-sources 0.8B document understanding model Xiaomi-OCR-0, supporting Markdown and JSON output
- 28 Oct, 08:00 · 1 reportsOvisOCR3 Technical ReportHugging Face Daily Papers:OvisOCR3 Technical Report
Coverage timeline
Follow the coverage to see every side of the story.
- ITHome · AIFeaturedXiaomi open-sources 0.8B document understanding model Xiaomi-OCR-0, supporting Markdown and JSON output
Xiaomi open-sources lightweight document understanding model Xiaomi-OCR-0, with 0.8B parameter size, weights and inference code have been uploaded to Hugging Face and GitHub, a single model covers three tasks of document parsing, key information extraction, and OCR visual question answering.
- Hugging Face Daily PapersOvisOCR3 Technical Report
OvisOCR3 is a 0.8B visual language model that unifies multiple OCR tasks through multi-mentor distillation and reinforcement learning, achieving SOTA results on document parsing, general OCR, key information extraction, image-text description, and cross-page table reconstruction tasks. It scored 96.74 on OmniDocBench v1.6, ranked first in OCRBench v2 English and Chinese split, and surpassed baseline models in multiple benchmark tests.