ocr / LANGUAGES.md
davanstrien's picture
davanstrien HF Staff
Sync from GitHub via hub-sync
3035c3c verified
|
Raw
History Blame Contribute Delete
5.35 kB

Language support

What each model's own card claims about language coverage — compiled from the model cards as of 2026-07-15. Treat these as claims, not guarantees: only one model (Surya) publishes per-language scores, and a "multilingual" tag tells you very little. Machine-readable version (plus params, backend, image pins): models.json.

The fastest way to find out whether a model handles your language is to run it on a few of your own pages — --max-samples 10 costs a few cents on a T4/L4:

hf jobs uv run --flavor l4x1 -s HF_TOKEN \
    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/surya-ocr.py \
    your-dataset your-output --max-samples 10

Sorted by strength of evidence, then coverage:

Model Card claim Evidence
Surya OCR 2 91 languages The only per-language benchmark published (full table): 38/91 score ≥90%, 76/91 ≥80%. Weakest of the top-15: Arabic 72.7%, Vietnamese 73.2%
dots.ocr 100 languages Explicit low-resource claim, backed by an in-house benchmark (1,493 PDFs across 100 languages) — aggregate scores only, not per-language
Tesseract 5 125 traineddata packs (~100+ languages + script models) Fully enumerable (tessdata_best) — the broadest named coverage here, incl. many low-resource languages; script models (Latin, Cyrillic, Devanagari, …) cover languages without a dedicated pack
Qianfan-OCR 192 languages Headline claim; no list or per-language numbers
PaddleOCR-VL / 1.5 / 1.6 109 languages; 1.5 adds Tibetan + Bengali Names scripts (Cyrillic, Arabic, Devanagari, Thai) and claims handwriting + historical documents; no per-language numbers
PP-OCRv6 48 languages Official card claim
Nanonets-OCR2-3B 11 named + "many more" (incl. Arabic + CJK) Illustrative list, no benchmark — but the only card claiming multilingual handwriting
LightOnOCR-2-1B 11 (9 European + zh/ja) Declared, not benchmarked. v1 is explicitly European/Latin-script only (9)
GLM-OCR 8 (zh en fr es ru de ja ko) Declared in card metadata only; no per-language evidence in the card body
HunyuanOCR-1.5 "Multilingual" Nothing enumerated, but explicitly targets low-resource + ancient-script OCR as a design goal (new in 1.5; the pinned 1.0 revision doesn't claim this)
dots.mocr · DeepSeek-OCR / -2 · Unlimited-OCR · NuExtract3 "Multilingual", unspecified A tag or one-liner only — no count, list, or benchmark
olmOCR-2 · Nanonets-OCR-s · SmolDocling · LFM2.5-VL-Extract English only Stated on the card
Falcon-OCR · OvisOCR2 · FireRed-OCR · ABot-OCR · RolmOCR · NuMarkdown-8B · lift Not stated No language information on the card at all

Text-only extraction: LFM2-1.2B-Extract (chains after OCR) names 9 languages: English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, Spanish.

Reading the table for low-resource work

  • Only Surya lets you check your language before running anything — its benchmark table has a row per language.
  • Tesseract's coverage is enumerable — if your language has a traineddata pack, it's supported (quality varies; it's the legacy baseline, not the quality leader).
  • dots.ocr and HunyuanOCR-1.5 are the VLMs that talk about low-resource languages at all; PaddleOCR-VL-1.5 names Tibetan and Bengali specifically.
  • A big claimed number (192, 109) without a list or per-language scores means you're testing it yourself either way — which is cheap (see the command above).