PaddleOCR-VL-1.6

PaddleOCR-VL-1.6

Reads an image and returns its text, built for parsing documents in English and Chinese. A 959M-parameter model from PaddlePaddle that runs through the PaddleOCR library.

Made by PaddlePaddle on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: PaddleOCR-VL-1.6 is an image-to-text model from PaddlePaddle for parsing documents. It has 959M parameters, ships as Safetensors files and plugs into the PaddleOCR library. Its card lists English and Chinese.

Who it suits: People who need the text out of scanned or photographed pages, and developers who already work with PaddleOCR.

Industry | Who uses it | How they use it | Example
Finance | Accounts teams | Turn scanned paperwork into text they can search | A scanned invoice becomes editable text
Legal | Paralegals | Pull the words out of scanned contracts | A signed PDF page becomes searchable text
Software | Developers | Add page reading to an app through PaddleOCR | An app reads a photographed form
Education | Researchers | Digitize printed papers and notes | A scanned article becomes text for analysis

You can skip it if your documents are already clean digital text, or if you need a language beyond English and Chinese.

Where to use it:

• Windows, macOS or Linux with Python, on your own computer
• Inside code that uses the PaddleOCR library
• Your own scripts and apps that read page images

Use case | Who | What they get back
Document parsing | Developers | Text read from a page image
Scan digitizing | Office staff | Text from scanned pages in English or Chinese
Quantized runs | People with small hardware | A smaller copy of the model, since others have published 27 quantized versions

How to set it up:

• Python on your computer
• The pip tool to install a package
• About 1.9 GB of disk for the BF16 weights, or about 479 MB at 4-bit as an estimate

1. Install the Hugging Face hub package.
2. Download the model files with the hf command.
3. Open the model card on Hugging Face and follow its steps to run the model.

pip install -U huggingface_hub
hf download PaddlePaddle/PaddleOCR-VL-1.6

How to use it: The run steps live in the model card on its Hugging Face page. After the download, follow them to point the model at a page image and read back the text.

hf download PaddlePaddle/PaddleOCR-VL-1.6

Example: You give it a scanned page and ask for the text on it. It answers with the words it read from the image, ready to copy into a document or a script.

For newcomers and experts: If you are new, start by downloading the files and opening the model card. Try one clean page image first, so you can check the output against what you see.

If you know the field, this is a vision-language model of 959M parameters in the PaddleOCR family, tuned for document parsing. The card title points to region refinement and staged post-training. Weights are small enough to quantize, and others have published 27 quantized and 16 fine-tuned versions.

Page image goes in
PaddleOCR-VL-1.6 reads the page
Text comes out

Good to know:

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the files you download to your computer and the images you give it
• Size: 959M parameters, about 1.9 GB in BF16

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: PaddleOCR
• Parameters: 959M
• Weights: about 1.9 GB in BF16; about 479 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 100,031
• Likes: 585
• License: apache-2.0
• Languages: en, zh
• Quantized versions: 27
• Fine-tunes: 16
• Tags: paddleocr, safetensors, paddleocr_vl, ernie4.5, paddlepaddle, image-to-text, document-parse, layout, table, formula, chart, seal

Links

• Model: https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6
• Files: https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download PaddlePaddle/PaddleOCR-VL-1.6

Then follow the usage in its model card. Nothing in this kit runs by itself.