Document Reader Powered by Xiaomi OCR 0
One lightweight 0.8B engine handles scans, spreadsheets and equations
None

None

None

Long Story Video Skill

Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill

Ads Video Skill

Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.

3D Science Explainer Video Skill

Convert scientific concepts into stunning 3D explain animations

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

xiaomi ocr 0

Scan any page with Xiaomi OCR 0 — a 0.8B vision-language model that returns tidy text, tables, math and handwriting in seconds.

All Tools

Discover our comprehensive AI-powered animation toolkit

Xiaomi OCR 0: Small Model, Serious Reading Skills

Rather than chaining separate detectors, Xiaomi OCR 0 handles layout reading, visual Q&A and field capture in one 0.8B checkpoint — replacing several narrow pipelines.

  • All-in-One 0.8B Checkpoint
    Layout reading, page-level Q&A and field capture all run inside the same lightweight network.
  • Benchmark-Leading Scores
    It ranks first on Real5-OmniDocBench with 95.24 and takes second on both OmniDocBench v1.6 and Wild-OmniDocBench.
  • Reliable on Messy Scans
    Skewed, curled, re-photographed or low-light pages are handled, so you are not limited to pristine PDFs.

Run Xiaomi OCR 0 in Three Quick Steps

From a single page image to tidy, structured results — no setup required.

Core Strengths of Xiaomi OCR 0

Compact size, dependable parsing, handwriting coverage and open weights: what Xiaomi OCR 0 actually delivers at 0.8B.

Page-Level Parsing

Works on ordinary PDFs as well as scanned, bent, tilted and re-photographed sheets.

Tables and Math

Formula recognition hits 97.84 on UniMER-Test, and table structure scores 92.27 on PubTabNet.

Handwritten and Calligraphic Text

Mature scripts reach 73 on Chronicles-OCR, matching Seed 2.0 Pro.

Page-Aware Question Answering

Averaging 83.2 across five VQA suites, it beats a 2B competitor while staying a fraction of the size.

Structured Field Capture

Receipts and forms come back as organized fields, not just a wall of plain text.

Open Weights, Local Runs

Apache-2.0 licensing with GGUF and BF16 builds ready for llama.cpp, Ollama and Transformers.

FAQ

Xiaomi OCR 0: Questions Answered

Straight answers on accuracy, handwriting, training data and running Xiaomi OCR 0 on your own machine.

1

What exactly is this model?

It is a 0.8B vision-language checkpoint tuned for reading documents: transcription and page-related questions happen in a single pass.

2

Where does it rank on benchmarks?

First on Real5-OmniDocBench, second on OmniDocBench v1.6 and on Wild-OmniDocBench.

3

Does handwriting work well?

Yes — clerical, cursive, running and regular calligraphy are all covered, scoring 73 on Chronicles-OCR.

4

What training went into it?

Around 170M OCR-focused samples, combined with Q-Mask text anchoring, continued pretraining and mixed-task reinforcement learning.

5

Can it run offline on my machine?

Yes. Apache-2.0 weights are public, and community GGUF and BF16 builds work with llama.cpp, Ollama and Transformers.

6

Is transcription the only thing it does?

No — it also answers questions about the page and pulls key fields into structured output.

Put Xiaomi OCR 0 to Work on Your Documents

Send one page image and get back readable text, rebuilt tables and organized fields — usually within moments.