[Models](/models)/xiaomi-research/Xiaomi-OCR-0

Apache 2.0

# Xiaomi-OCR-0

by xiaomi-research

Xiaomi's 0.8B OCR model that parses documents and answers questions about them, including charts, tables and formulas, under Apache 2.0.

177likes

0.8Bparams

[HuggingFace](https://huggingface.co/xiaomi-research/Xiaomi-OCR-0)[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

Identifiers

Model ID

`xiaomi-research/Xiaomi-OCR-0`

## Deploy Xiaomi-OCR-0

Single-tenant

Mixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.

[Read the customer-hosted guide](https://mixpeek.com/docs/customer-hosted/overview) [Talk to us about single-tenant](/demo)

## Overview

Xiaomi-OCR-0 is a 0.8B-parameter vision-language model from Xiaomi, released on September 30, 2026. One model handles whole-page parsing, OCR-oriented visual question answering and key information extraction, so it can return a page as Markdown or answer a question about what the page shows. Xiaomi reports 95.24 on Real5-OmniDocBench, where it ranks first on its own comparison table, and 84.6 on the ChartQA benchmark. Weights are on Hugging Face under Apache 2.0.

## Architecture

Built from Qwen3.5-0.8B-Base and trained on a corpus of about 170M OCR samples. Xiaomi describes three training steps: Q-Mask text anchoring to align fine-grained text with its location, OCR-centric continued pretraining, and mixed-task reinforcement learning with verifiable rewards on hard examples.

## Mixpeek SDK Integration

IngestRetrieve

\# Parse one document image to Markdown (adapted from the Xiaomi-OCR-0 model card).
# pip install "transformers==5.17.0" "torch==2.14.0" "torchvision==0.29.0" pillow
import torch
from transformers import AutoModelForImageTextToText, AutoProcessor

MODEL = "xiaomi-research/Xiaomi-OCR-0"
PROMPT = (
    "Extract all information from the main body of the document image and represent it "
    "in markdown format, ignoring headers and footers. Tables should be expressed in OTSL "
    "format, formulas in the document should be represented using LATEX format, and the "
    "parsing should be organized according to the reading order."
)

processor = AutoProcessor.from\_pretrained(MODEL)
model = AutoModelForImageTextToText.from\_pretrained(MODEL, dtype=torch.float16).to("cuda").eval()

messages = \[{"role": "user", "content": \[
    {"type": "image", "url": "page.png"},
    {"type": "text", "text": PROMPT},
\]}\]
inputs = processor.apply\_chat\_template(
    messages, tokenize=True, add\_generation\_prompt=True,
    return\_dict=True, return\_tensors="pt",
).to("cuda")

with torch.inference\_mode():
    output = model.generate(\*\*inputs, max\_new\_tokens=4096, do\_sample=False)
print(processor.batch\_decode(output\[:, inputs\["input\_ids"\].shape\[1\]:\], skip\_special\_tokens=True)\[0\])

## Capabilities

*   Whole-page parsing to Markdown, with tables in OTSL and formulas in LaTeX
*   Question answering over documents, infographics and charts
*   Key information extraction to a JSON schema you supply
*   Scanned, warped, photographed and skewed pages
*   GGUF and Ollama community builds, plus Transformers

## Use Cases on Mixpeek

Parsing scanned PDFs and phone photos into Markdown before chunking them for RAG

Answering a question about a chart or infographic on a page without a separate OCR and LLM step

Pulling named fields from invoices and forms by passing a JSON schema

Reading the text inside images for alt text and search

## Benchmarks

Dataset

Metric

Score

Source

Real5-OmniDocBench

Overall score

95.24

https://huggingface.co/xiaomi-research/Xiaomi-OCR-0

OmniDocBench v1.6

Overall score

96.83

https://huggingface.co/xiaomi-research/Xiaomi-OCR-0

ChartQA

Score

84.6

https://huggingface.co/xiaomi-research/Xiaomi-OCR-0

## Performance

Input SizeDocument images and PDF pages

GPU LatencyInput dependent

GPU ThroughputBatch dependent

GPU MemoryModel dependent

We have not measured it. Scores are Xiaomi's own, from its model card.

## Common Pipeline Companions

[

openbmb/MiniCPM-OCR-1.0

A 1.3B document parser under the same license

](/model/openbmb/MiniCPM-OCR-1.0)[

PaddlePaddle/PaddleOCR-VL-1.6

A 0.9B document parser Xiaomi compares against

](/model/PaddlePaddle/PaddleOCR-VL-1.6)[

baidu/Unlimited-OCR-2

A larger OCR model built for long documents

](/model/baidu/Unlimited-OCR-2)

## Frequently Asked Questions

### What is Xiaomi-OCR-0?

A 0.8B-parameter OCR model from Xiaomi that parses documents and also answers questions about them. It covers Markdown parsing, tables, formulas, charts and field extraction with a single model.

### Can Xiaomi-OCR-0 read charts?

Xiaomi reports 84.6 on the ChartQA benchmark, ahead of the 0.8B, 2B and 4B Qwen3.5 models in its table and behind MiniCPM-V-4.5 at 87.4. Treat values it reads from a chart as needing a check against the source page.

### Can I use Xiaomi-OCR-0 commercially?

The model card lists Apache 2.0.

### What output formats does it return?

Markdown for whole pages, OTSL for tables, LaTeX for formulas, plain text for text regions, and a JSON object for key information extraction when you pass a schema.

### How do I use it with Mixpeek?

Run it where your documents are, store each page's Markdown and fields with the file and page number, and upsert them into a collection so the text is searchable and the fields filterable.

## Explore on Mixpeek

[

Browse all models

Search the full model catalog

](/models)[

Feature Extractors

See how Mixpeek runs models as extractors

](/extractors)[

Mixpeek Vector Store

Store & search embeddings at scale

](/mvs)[

Pricing

Usage-based pricing for pipelines

](/pricing)[

All Curated Lists

Compare models, APIs & infrastructure

](/curated-lists)

### Specification

Organizationxiaomi-research

Retriever\-

Parameters0.8B

LicenseApache 2.0

Downloads/moN/A

Likes177

### Build a pipeline with Xiaomi-OCR-0

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)