# Xiaomi Enters Document OCR With New Vision Model

> The company's research arm releases Xiaomi-OCR-0, a document-parsing vision-language model built on a Qwen3.5 backbone.

Published by The Open Weights on Sep 29, 2026. Canonical: https://theopenweights.com/news/xiaomi-ocr-0-87qa

## Key facts

- Company: Xiaomi
- Model: Xiaomi-OCR-0
- Version: 0
- Category: Vision-Language
- Modalities: Vision-Language
- License: Other (Open weights, commercial use allowed)
- Context window: 262K tokens
- Architecture: Qwen3_5ForConditionalGeneration
- Precision: BF16
- Weights size: 1.7 GB
- Significance: notable
- Published: 2026-09-29
- Last verified: 2026-11-01
- Hugging Face: https://huggingface.co/xiaomi-research/Xiaomi-OCR-0
- Canonical URL: https://theopenweights.com/news/xiaomi-ocr-0-87qa

Xiaomi's research group has published [Xiaomi-OCR-0](https://huggingface.co/xiaomi-research/Xiaomi-OCR-0), a vision-language model aimed squarely at document understanding. According to the release, the model handles optical character recognition and structured document parsing, and is built on a Qwen3.5 backbone — reflecting the growing practice of adapting strong open foundation models rather than training from scratch.

Document OCR has become one of the most practical frontiers for multimodal AI. Beyond simply reading text from images, modern systems are expected to preserve layout, extract tables, and output structured representations that downstream software can consume. That makes a dedicated parsing model more useful than a general-purpose chatbot for enterprises digitizing invoices, forms, and reports.

## Why it matters

The release signals Xiaomi's continued investment in applied AI research and adds another option to a field that already includes capable open OCR models from other labs. A few points stand out:

- It targets both **OCR and document parsing**, not just raw text extraction.
- It leverages the **Qwen3.5** family as its foundation.
- It ships under a custom license, so teams should review the terms before commercial use.

As an initial "0" release, Xiaomi-OCR-0 reads more like a starting point than a finished product line. Key details such as parameter count and context length are not specified in the listing, so practitioners will want to test it against their own document workloads before committing. The model and its documentation are available now on [Hugging Face](https://huggingface.co/xiaomi-research/Xiaomi-OCR-0).

## Get the model

- [Hugging Face](https://huggingface.co/xiaomi-research/Xiaomi-OCR-0)

## Sources

- [xiaomi-research/Xiaomi-OCR-0](https://huggingface.co/xiaomi-research/Xiaomi-OCR-0) — Hugging Face, Sep 29, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "Xiaomi Enters Document OCR With New Vision Model", The Open Weights, Sep 29, 2026, https://theopenweights.com/news/xiaomi-ocr-0-87qa