CAICHINAAI

← Vision

Multimodal / Vision

Xiaomi-OCR-0

Xiaomi-OCR-0 by Xiaomi Research has no tracked price. It has no overseas route listed on this record. Last checked 2026-11-01.

Xiaomi Research · apache-2.0 · 0.8B

—

unpriced · Last updated · Pricing source

Document-understanding model for parsing, key-information extraction, and OCR visual QA. Model card: 0.8B, continued from Qwen3.5-0.8B-Base. HF safetensors count is 0.873B. Apache-2.0. config.json text_config.max_position_embeddings is 262144; the card does not state a context window. Model card news line dates the weights release 2026-09-30. arXiv 2609.36136 was submitted 2026-09-28. GitHub org Xiaomi Research. No USD price.

How to use it from outside China

  1. Weights are public. Download them from Hugging Face. https://huggingface.co/xiaomi-research/Xiaomi-OCR-0
  2. This record does not list an official international API, OpenRouter, or another overseas host.
  3. Confirm the price on the source before you rely on it. Checked 2026-11-01. Pricing source

Usable outside China?

No official international API listed.

Not listed on OpenRouter.

Questions

Can I use Xiaomi-OCR-0 from outside China?

Unknown — not on the record yet

How much does Xiaomi-OCR-0 cost per 1M tokens?

Unknown — not on the record yet

Is the Xiaomi-OCR-0 API OpenAI-compatible?

Unknown — not on the record yet

Are Xiaomi-OCR-0 weights open, and under which license?

Yes. License: apache-2.0.

Sources

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

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

A dash means we don’t know. This page does not fill gaps.

Related models

Other vision models

Compare Xiaomi-OCR-0 with DeepSeek V4.1 Flash 1017 · More vision · Xiaomi Research

Weekly brief

Get the weekly brief on Chinese AI models you can use outside China. Free.