typhoon-ai/typhoon-ocr1.5-2b

Image-Text-to-Text • 2B • Updated Jun 11 • 192k • • 28

Note Thai/English page parsing, forms and handwriting; use the required OCR prompt and Typhoon tooling. Usage: https://github.com/scb-10x/typhoon-ocr

sbintuitions/sarashina2.2-ocr

Image-to-Text • 4B • Updated Mar 31 • 1.55k • 33

Note Japanese/English pages to Markdown, including vertical Japanese text, HTML tables and LaTeX formulas.

5CD-AI/Vintern-1B-v3_5

Image-Text-to-Text • 0.9B • Updated Dec 10, 2025 • 7.09k • 125

Note Vietnamese forms, invoices and text-rich images; use explicit transcription prompts with this broader document VLM.

NAMAA-Space/Qari-OCR-0.4.0-VL-4B-Instruct

Image-to-Text • Updated Feb 1 • 681 • 7

Note Adapter for printed Arabic Islamic books; requires its Qwen3-VL base model, with less established support for handwriting and modern fonts.

NCSOFT/VARCO-VISION-2.0-1.7B-OCR

Image-Text-to-Text • 2B • Updated Sep 15, 2025 • 1.13k • 29

Note Korean/English recognition with character bounding boxes; non-commercial licence.

wayu-ai/wayu-paxa-ocr-zero

Image-Text-to-Text • 0.9B • Updated 3 days ago • 35 • 1

Note Thai/English region recognition; whole pages need layout detection, and this research release has no maintenance commitment. Usage: https://github.com/wayu-research/wayu-ocr-inference

PaddlePaddle/arabic_PP-OCRv5_mobile_rec

Image-to-Text • Updated Oct 16, 2025 • 1.17k • 7

Note Arabic-script line recognition in PaddleOCR; supply text crops or add detection for full pages, and check coverage on your fonts.

PaddlePaddle/devanagari_PP-OCRv5_mobile_rec

Image-to-Text • Updated Oct 16, 2025 • 1.47k

Note Devanagari line recognition with Hindi examples; crop lines or use PaddleOCR detection, and check your language’s coverage.

PaddlePaddle/korean_PP-OCRv5_mobile_rec

Image-to-Text • Updated Jul 22 • 5.38k • 17

Note Dedicated Korean line recognition with PaddleOCR examples; use text crops or the full detection pipeline.

kha-white/manga-ocr-base

Image-to-Text • Updated Jun 22, 2022 • 889k • 180

Note Japanese manga text, including vertical writing and furigana; supply one text region, since blank images can produce invented text. Usage: https://github.com/kha-white/manga-ocr