China Hot Tech @ChinaHotTech
← All posts
Models · OCR · Open Source

Xiaomi-OCR-0: 0.8B Open Unified Document OCR / Understanding VLM

Xiaomi-OCR-0 is Xiaomi’s open 0.8B vision-language model that unifies document parsing, key information extraction (KIE), and OCR-centric VQA in one instruction interface. It starts from Qwen3.5-0.8B-Base, trains on roughly 170M OCR-centric samples, and ships under Apache-2.0 on Hugging Face with a Transformers path. Chinese coverage (智东西, 2026-10-31) ties the stack to Super Xiaoai screen-understanding on Xiaomi phones — but what overseas engineers can actually pull today is the model weights and code, not a new handset SKU. Vendor and paper benches below are flagged with asterisks.

· ~7 min read · China Hot Tech

Soft paper-white and coral editorial illustration: phone and scanned document dissolving into Markdown and JSON structured fields — Xiaomi-OCR-0 OCR to understanding

Illustration: paper-white / coral / ink — pixels dissolving into Markdown and JSON. Distinct from recent mint/cream and chip-cluster art. Asterisked labels are vendor/paper figures.

Compact OCR VLMs have gotten good at turning a page into text. The harder product ask — for phone assistants, agent tools, and RAG pipelines — is turning a messy photo of a page into structure you can act on: reading order, nested tables, formulas, and schema-shaped fields. Xiaomi-OCR-0’s pitch is that one 0.8B model can cover that stack without splitting “parser” and “understander” into separate products. For overseas readers the concrete checks are license, install path, how the headline benches were measured, and whether community GGUF/Ollama packs match the paper numbers.

What the model actually does

On the Hugging Face card and technical report, Xiaomi-OCR-0 is framed as a unified OCR-specialized VLM:

Training is progressive: Q-Mask text anchoring for fine-grained location↔token alignment, then OCR-centric continued pretraining (CPT), then Mix-RL on hard examples with verifiable rewards (edit similarity, TEDS, CDM, ANLS, JSON field match). The data engine leans on heterogeneous-expert consensus, render-guided verification, and targeted synthesis across that ~170M-sample corpus.

Can you use it today?

As weights + Transformers inference — yes: huggingface.co/xiaomi-research/Xiaomi-OCR-0 (Apache-2.0). The card documents a verified transformers / torch combo and AutoModelForImageTextToText whole-page parsing prompts. Code, batch scripts, two-stage layout→region assembly, Agent Skill, and eval helpers live on GitHub xiaomi-research/Xiaomi-OCR-0. Paper: arXiv 2609.36136. Community GGUF / Ollama / NInfer packs exist (HF news cites several); Xiaomi explicitly has not evaluated those conversions for accuracy. Browser demo / MCP / Agent Skill are mentioned on the card and in 智东西 coverage. As a phone you buy — no. Super Xiaoai screen-understanding is the product narrative in Chinese coverage; the overseas deliverable is open weights.

Three-panel editorial: document parsing, KIE, OCR-VQA with training recipe Q-Mask → CPT → Mix-RL and overseas HF / Transformers path

Parse · KIE · OCR-VQA on one 0.8B stack; Q-Mask → CPT → Mix-RL. Asterisks mark vendor/paper benches; main OmniDoc numbers use a two-stage layout detector.

Headline benches* (vendor / paper)

Prefer the HF model card and arXiv report over secondary blogs for the numbers. Selected figures:

Read the protocol carefully: the report states that the main OmniDocBench v1.6 / Real5 / Wild comparisons use two-stage inference — PP-DocLayoutV3 detects regions, Xiaomi-OCR-0 recognizes crops, then pages are assembled. The advertised 0.8B count excludes that external layout detector. Formula CDM is a highlighted strength on OmniDocBench*; OvisOCR2 and TeleOCR still win some text-edit / table cells in the same tables*.

What changes for an overseas stack

If you already run PaddleOCR-VL, MinerU, GLM-OCR, dots.ocr, or a Qwen3.5-VL baseline for docs, Xiaomi-OCR-0 changes three practical checks:

Caveats

Bottom line: Xiaomi-OCR-0 is a compact, openly licensed bet that document structure and OCR-centric understanding can share one 0.8B instruction model. Overseas teams can pull HF weights today, follow the Transformers path, and keep asterisks on every leaderboard claim — especially wherever a separate layout detector is in the loop.

Sources