Reportedly: Xiaomi open-sources Xiaomi-OCR-0, a 0.8B-parameter document understanding model
Model: Xiaomi-OCR-0 · availability, license and releases
Reported by IT Home · not yet confirmed by the company or a second independent outlet. We update this page when it is.
Xiaomi released Xiaomi-OCR-0, a 0.8B-parameter open-source model for document parsing, key information extraction, and OCR visual question answering. Weights and inference code are on Hugging Face and GitHub. It is built on Qwen3.5-0.8B-Base and trained on about 170 million OCR and document understanding samples, IT Home reported.
China context
- Original name
- 小米
- Outside China
- Open weights · ithome.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Builders can integrate Xiaomi-OCR-0 into document processing pipelines for local OCR and structured data extraction, with support for Markdown, LaTeX, and JSON outputs.
- For investors
- Investors should monitor adoption of Xiaomi-OCR-0 in enterprise document automation and whether it pressures commercial OCR API pricing.
Translated from Chinese. Quotes and facts link to the original sources.
Xiaomi announced the open-source release of Xiaomi-OCR-0, a lightweight document understanding model with 0.8 billion parameters. The model handles document parsing (recognizing text, tables, and math formulas with structured output in Markdown and LaTeX), key information extraction (from invoices, receipts, cards, and business forms, outputting JSON), and OCR visual question answering (cross-region retrieval, numerical calculation, and semantic reasoning over documents and charts). It is based on Qwen3.5-0.8B-Base and trained on approximately 170 million OCR and document understanding samples, using a heterogeneous expert consensus mechanism for data annotation.
The model uses a heterogeneous expert consensus mechanism where multiple OCR expert models cross-validate and vote to select high-confidence annotations for different regions (text, tables, formulas). This approach aims to improve data quality for training. The model supports structured outputs in Markdown, LaTeX, and JSON, enabling integration into automated pipelines.
Developers outside China gain a small, open-weight OCR model that can run locally for document parsing and structured data extraction, reducing reliance on cloud OCR APIs and lowering per-document processing costs.
For businesses, this model offers a cost-effective option for automating document processing tasks such as invoice parsing and form data extraction, with the flexibility of open weights for customization and on-premise deployment.
Next signals to check: independent benchmarks comparing Xiaomi-OCR-0 against other open OCR models on document parsing accuracy, and whether the model is adopted in downstream applications or fine-tuned for specific domains.