Hebrew OCR

Image to Text Hebrew — Extract RTL Hebrew Script with OCR

Convert Hebrew image to text on ExactRead: upload a PNG, JPEG, or WebP containing right-to-left Hebrew script and compare OCR models — set a language hint for accurate RTL extraction, export TXT or JSON.

Open workbench View pricing

Try OCR now

Upload a document and compare models right here — no need to leave this page.

Hebrew OCR and right-to-left reading order

Hebrew is a right-to-left script with 22 consonant letters and optional vowel diacritics (nikud). Most printed Hebrew omits nikud entirely, relying on readers to infer vowels from context — which means OCR engines must handle consonant-only text without the vowel anchors that help with other scripts. An additional challenge is bidirectionality: Hebrew documents regularly embed Arabic numerals, Latin product names, or English phrases read left to right, all within a right-to-left paragraph. Setting the language hint to Hebrew (he) before running image to text tells the model to treat reading order as right-to-left and prioritize Hebrew character recognition.

Hebrew use cases: press, legal, and product labels

Hebrew image to text is used for Israeli newspapers and magazine scans where Hebrew text wraps around embedded Latin headlines, legal contracts and court documents in Hebrew, product packaging labels that combine Hebrew product names with Latin ingredient lists, government-issued ID and medical documents, and historical archive scans where older typefaces and low-contrast paper require careful model selection. Each scenario has different font and layout characteristics — compare several models on a representative sample before standardizing on one engine.

Models for Hebrew image to text

GPT-5.4 and Gemini 3 Flash handle Hebrew in vision mode, including bidirectional documents that mix Hebrew and Latin. Azure Document Intelligence includes explicit Hebrew language support and handles right-to-left layout in structured documents. Google Cloud Vision OCR supports Hebrew (he) as a language hint. For documents where precise RTL column order matters — legal forms, financial statements — compare a native OCR engine with a vision model and cross-check the reading order in the output.

Review Hebrew OCR output carefully

The most common Hebrew OCR errors are letter substitutions between visually similar pairs — bet/kaf, dalet/resh, vav/zayin — and dropped or misplaced vowel points when nikud is present. ExactRead displays confidence and warning signals alongside each result, which helps reviewers focus on the uncertain lines rather than reading the full output twice. Export accepted results as TXT for direct use in Hebrew-aware editors, or as structured JSON when the content feeds a database or a translation pipeline that needs full text, confidence, and warnings together.

How it works

1
Upload your Hebrew document
Drop a Hebrew document onto the workbench or click to browse. Supported inputs are PNG, JPEG, and WebP images, plus PDF for PDF-capable models.
2
Let ExactRead filter the models
After upload, the model list is filtered to the OCR engines that actually support your file, so you never start a job that cannot run.
3
Run one model or compare several
Choose a single model when speed and cost matter, or compare mode to run several OCR models on the same Hebrew document at once.
4
Review confidence and warnings
Each result keeps its model, status, confidence, and warnings, with the recommended output highlighted so you can judge accuracy quickly.
5
Accept and export TXT or JSON
Accept the output that reads your Hebrew document most faithfully, then copy it or export TXT or JSON for the next step.

Models that suit this document type

Capability, format support, and credits come straight from the model catalog — a starting point, not a ranking. Test on your own documents to decide.

Model	Provider	Type	Formats	Credits / doc
GPT-5.4	OpenAI	Vision-language model	Image files	3
Gemini 3 Flash	Google	Vision-language model	Image files	1
Azure Document Intelligence	Microsoft Azure	Native OCR engine	Images and PDF	2
Google Cloud Vision OCR	Google	Native OCR engine	Images and PDF	2

Supported formats

PNG, JPEG, WebP — right-to-left Hebrew documents

Model guidance

GPT-5.4, Gemini 3 Flash, Azure, Google Vision; set language hint to he

Credit note

Free plan: 100 credits/month; compare native OCR + vision for RTL

Back to ExactRead home View OCR credit pricing

Frequently asked questions

Which models support Hebrew OCR?

GPT-5.4 and Gemini 3 Flash handle Hebrew in vision mode, including RTL layout. Azure Document Intelligence and Google Cloud Vision OCR have explicit Hebrew language support. Set the language hint to he for best results.

Does Hebrew OCR preserve right-to-left order?

Vision models and native OCR engines with Hebrew language support generally preserve RTL reading order. For critical documents, always review the plain text output against the original image to confirm paragraph and column order is correct.

Can it handle mixed Hebrew and English text?

Yes. Vision models like GPT-5.4 and Gemini 3 Flash handle bidirectional Hebrew-English content. Set the language hint to auto so the model manages both scripts without being forced to treat the whole image as a single language.

Does it handle vowel points (nikud) in Hebrew?

Modern printed Hebrew usually omits nikud. When nikud is present — in religious texts, children's books, or teaching materials — vision models perform better than pattern-based OCR because they can infer the vowel mark from context. Review the output carefully, as nikud substitution errors are common.

Is Hebrew image to text free to try?

Yes. The free plan includes 100 OCR credits every month on sign-in — enough to compare several models on a set of Hebrew images and pick the engine that reads your documents best.