Truly multilingual
Recognizes 100+ languages including CJK, Arabic, and Cyrillic scripts.
PaddleOCR-VL 1.6
A compact multilingual VLM that parses text, tables, formulas, and handwriting across 100+ languages.
Price$0.001 / page
ProviderPaddlePaddle
Parameters~0.9B parameters
ApproachUnified detection + recognition VLM
Languages100+ languages
LicenseApache 2.0
ReleasePaddleOCR-VL v1.6 · 2026
Overview
What it does
PaddleOCR-VL 1.6 is Baidu’s lightweight vision-language OCR model from the PaddlePaddle ecosystem. It unifies text recognition, layout analysis, table parsing, and formula extraction in a single sub-billion-parameter model with exceptionally broad language coverage.
Designed for the real world, it handles printed text, handwriting, rotated scans, and noisy photos with ease. Backed by the battle-tested PaddleOCR toolkit, it’s a dependable all-rounder for global, multilingual document pipelines.
Best for
- Global, multilingual document processing
- Handwriting and form recognition
- Edge and on-device OCR deployments
Benchmarks
How it measures up
olmOCR-Bench
81.0
overall accuracy score
Languages
100+
scripts supported
Params
0.9B
edge-friendly
Throughput
4.6 pg/s
on a single H100
Figures are representative of the PaddleOCR-VL 1.6 model card; see Hugging Face for the full evaluation suite.
Highlights
Handwriting & photos
Robust to handwriting, rotation, and the noise of real-world captures.
Formulas & tables
Extracts math to LaTeX and reconstructs tables in one unified pass.
