Bayanat Knowledge

Arabic text and knowledge for model training.

License books, subject-organized MSA text, and structured question-answer data. Each collection has a clear training task, with its own counts and licensing details.

Browse datasets ↓Need help choosing?

Find your text dataset

Choose a collection to see its counts, intended use, and licensing details.

2,700 books

Arabic Books

For model teams that need longer Arabic texts for language modeling or deeper coverage of a particular subject.

View dataset 400,000 multiple-choice Q&A items

Arabic STEM Q&A

For teams training Arabic educational assistants or models that answer STEM questions.

View dataset 200,000 sentences

MSA Named Entity Recognition

For teams building Arabic entity extraction into search, document processing, or language-understanding systems.

View dataset 180,000 question-answer pairs

Arabic Regional Q&A

For teams building Arabic assistants whose users ask questions about the Arab region.

View dataset 40,000 question-answer pairs

English Regional Q&A

For teams building English-language assistants that need to answer questions about the Arab region.

View dataset Ask about available volume

Edited MSA Knowledge

For model teams adapting an Arabic model to a subject where edited language and clear content rights matter.

View dataset

Which Arabic text dataset fits your task?

Start from what your model needs to learn. Ready collections are licensed for the selected titles and uses; the last three rows are jobs that belong to custom collection or to another Bayanat family.

Your task	Start with	Check before licensing
Long-form Arabic for domain adaptation or continued pretraining	Arabic Books (2,700 books)	The title, publisher and subject list, and any token count with its tokenizer named and raw and cleaned text counted separately.
Clean Modern Standard Arabic organized by subject, with commercial AI-training rights in the agreement	Edited MSA Knowledge	Subject coverage and sample records. It covers material through 2026, with commercial AI-training rights in the agreement; retrieval, display and redistribution are scoped separately.
STEM multiple-choice items for training	Arabic STEM Q&A (400,000 items)	Answer keys, whether explanations are included, and any proposed training/test split. For an independent test set, use Bench instead.
Regional knowledge for an assistant	Arabic Regional Q&A (180,000 pairs) English Regional Q&A (40,000 pairs)	Field lists and answer review. The two are separate products; do not assume they are parallel translations.
Entity extraction from formal Arabic text	MSA Named Entity Recognition (200,000 sentences)	Entity types, nesting, tokenization and export format.
Dialectal or mixed Arabic-English text at volume	Custom collection	Request the register and dialect breakdown for the ready collection you are considering. If it does not cover what you need, scope a collection from authorized sources.
Passages with relevance judgments for search or RAG	Bayanat Search Arabic Retrieval Starter Kit (free)	Relevance scale, hard negatives and unanswerable questions; the free Arabic Retrieval Starter Kit shows the format.
Text inside scanned pages, forms or photos	Bayanat Documents 10 checks for Arabic OCR data	Transcription, reading order and field evidence; the 10-point Arabic OCR audit lists what to inspect.

Is Bayanat Knowledge the right fit?

A good fit if

You are adapting a model to Arabic subject matter and need licensed long-form or professionally edited text.
You need structured Arabic question-answer data or STEM multiple-choice items for training.
You need Modern Standard Arabic entity annotations for extraction.

Choose something else if

You need an independent test set. Items you license for training are not a held-out benchmark; see Bench.
You need dialectal or code-switched text at volume that the collections’ register and dialect breakdowns do not show. Commission a custom collection.

What you can check before you license

We separate what is confirmed on this page, what we send on request, and what is agreed for each project.

Item	Status	Details
Counts per collection	Confirmed	2,700 Arabic books; 400,000 Arabic STEM multiple-choice items; 200,000 MSA NER sentences; 180,000 Arabic regional Q&A pairs; 40,000 English regional Q&A pairs (a separate product; parallel translation is not established).
Edited MSA coverage and rights	Confirmed	Professionally edited, subject-organized Modern Standard Arabic covering material through 2026, with commercial AI-training rights in the agreement.
Title, publisher and subject lists	On request	For books and edited MSA, before you choose material.
Token counts	On request	Any token count names the tokenizer and separates raw from cleaned text.
NER label taxonomy and span rules	On request	Entity types, nesting, tokenization and export format.
Answer keys, explanations and splits	On request	Field lists, answer review and proposed training/test boundaries.
Sample records	On request	Ask for them with the specifications for the collection you select.

Need more than the existing collection?

Our team can add data or annotation for your particular model. These additions are agreed separately.

Books and edited text

Choose Arabic books or professionally edited MSA organized by subject. The edited collection covers material through 2026.

Question-answer data

Arabic STEM multiple-choice questions sit alongside separate Arabic and English Q&A collections about the Arab region.

Entity annotation

The 200,000-sentence MSA named-entity-recognition collection supports extraction from formal Arabic text.

Additional subjects

Tell us which professional, educational, technical, or regional material you need. We can plan authorized sourcing and annotation around that need.

How the license covers each use

Training rights do not automatically include display, retrieval or redistribution. Each use is set in the agreement.

Model training

Commercial license for the selected titles and uses. The edited MSA collection has commercial AI-training rights written into the agreement.

Retrieval and display

A separate licensed use. Tell us if your product shows passages to users or retrieves them at runtime.

Redistribution

A separate scope. Specify it if you need it.

How we put the data together

Find the right material

Review a sample, subject coverage, register, date coverage, and the unit used to report volume.

Plan the format

Custom preparation can include document boundaries, source records, publication dates, license identifiers, cleaning history, and deduplication records.

Confirm the rights

The edited MSA collection includes commercial AI-training rights in the agreement. Training, retrieval, display, and redistribution are separate licensed uses.

For example: Building a technical Arabic assistant

Start with edited material in the relevant subject, add questions tied to that material, and use a separate evaluation set to check the assistant’s answers.

Scope your request

Send us these details and we’ll send the available specifications, and a sample where one is available, so your team can check the fit before licensing.

Collection(s) you are interested in
Subjects or domain
Intended use: training, retrieval or display
Volume needed
Timeline

Request collection specifications and samples

Common questions

Are the Arabic and English regional Q&A sets translations of each other?

They are separate products: 180,000 Arabic pairs and 40,000 English pairs. Parallel translations or item-level alignment are not assumed; request the collection specifications if your application needs matched pairs.

What rights come with the edited MSA collection?

Commercial AI-training rights are written into the agreement. If you also need to display passages, use the material for retrieval, or redistribute it, those uses need to be specified separately.