Bayanat Labs Open: Free Arabic Dataset Resources
Every free Arabic dataset, tool and benchmark published by Bayanat Labs, with licence, DOI and citation (updated 19 October 2026). اقرأ بالعربية ←
In the catalogue
Filter releases
No release matches these filters.
Releases
Bayanat Labs Arabic AI Model Tracker
89 Arabic and Arabic-capable AI model releases from 29 labs: date, sizes, Arabic varieties named, lab-reported scores, licence and weights status.
Bayanat Labs Dataset Licence Compatibility Checker
Check whether dataset sources can be merged, trained on or redistributed, with the obligations, release licence options and the clause behind each step.
Bayanat Labs Arabic Document Atlas
200,286 unique document images with machine OCR, approximate word guides and source evidence, plus separate community text and PDF/reference pairs.
Bayanat Labs Arabic Annotation Guidelines Template
Free Arabic annotation guidelines template under CC BY 4.0: sections for dialect, diacritics, Arabizi, examples and pilots. Licence, citation and reuse terms.
Bayanat Labs Arabic ASR Channel Pairs
Free Arabic speech test set: FLEURS recordings, clean and under simulated telephone, codec and room conditions, in aligned pairs, with Whisper WER.
Bayanat Labs Arabic Text Normalizer (arabic-normalize)
Release record for arabic-normalize: named, versioned Arabic normalisation policies in Python and JavaScript, with conformance fixtures. Apache-2.0 code.
Bayanat Labs Arabic PII Pack for Presidio
Release record for arabic-pii-pack: Presidio recognisers and redaction for Gulf and Egyptian IDs, Arab IBANs and phones, plates and Arabic names.
Bayanat Labs Arabic Token-Tax Index
How many more tokens Arabic takes than English: LLM tokenisers measured on UN parallel text, with bootstrap intervals and the cost of a page.
Bayanat Labs Arabic Unicode Conformance Suite
Release record for the Arabic Unicode Conformance Suite: test cases with exact expected outputs under six named normalisation policies, and a runner.
Bayanat Labs ArabicWER: Arabic WER and CER Scorer
Release record for ArabicWER: Arabic WER and CER under six named, versioned normalisation policies, with conformance fixtures and a FLEURS case study.
Bayanat Labs Contributor Consent Templates
Free consent form, contributor agreement, privacy notice and more for collecting open Arabic data under the Saudi PDPL, in English and Arabic. CC BY 4.0.
Bayanat Labs Dataset Template and Release Kit
Release record for the Bayanat Labs dataset template: a copier template, a bilingual dataset-card standard and a release kit with a card linter and gates.
Bayanat Labs GCC Format Validators and Arabic ITN/TN Test Pairs
Release record: Arabic spoken and written test pairs for ITN and TN, with Gulf format tables and validators for IDs, IBANs, phones, plates and Hijri dates.
Bayanat Labs Open-Programme Policy and Operating Manual
Release record for the Bayanat Labs open policies: the open-programme policy, the gated operating manual and the bl-score tool, in English and Arabic.
Bayanat Labs Arabic Agent Reliability Lab
Arabic Agent Reliability Lab release record: a local harness that grades Arabic AI agents by tool calls and final state. Apache-2.0 code, CC BY 4.0 data.
Bayanat Labs Arabic Retrieval Starter Kit
Download the free Arabic Retrieval Starter Kit v0.1: a synthetic Arabic and English RAG evaluation fixture under CC BY 4.0, with qrels, DOI and citation.
Bayanat Labs LabelBench
LabelBench release record: aggregate Arabic labeling results (CC BY 4.0), Apache-2.0 code on GitHub and PyPI, Zenodo DOI, citation and licences.
All releases at a glance
| Release | Type | Licence | Size | Version | DOI | Updated |
|---|---|---|---|---|---|---|
| Arabic AI Model Tracker | Dataset | CC BY 4.0 | 89 model releases recorded | v0.1.0 | None | |
| Dataset Licence Compatibility Checker | Tool | Apache-2.0 | 33 licences and terms encoded | v0.1.0 | None | |
| Arabic Document Atlas | Dataset | Component-specific rights | 200286 unique decoded document images across image components | v0.3.0 | None | |
| Arabic Annotation Guidelines Template | Guide | CC BY 4.0 | 12 copy-ready template sections | Unversioned | None | |
| Arabic ASR Channel Pairs | Dataset | CC BY 4.0 | 500 clean FLEURS recordings (parents) | v0.1.0 | None | |
| Arabic Text Normalizer (arabic-normalize) | Tool | Apache-2.0 | 80 conformance fixtures | v0.1.0 | None | |
| Arabic PII Pack for Presidio (arabic-pii-pack) | Tool | Apache-2.0 | 1000 synthetic test sentences (every value generated) | v0.1.0 | None | |
| Arabic Token-Tax Index | Benchmark | CC BY 4.0 | 12 tokenisers measured | v0.1.0 | None | |
| Arabic Unicode Conformance Suite | Benchmark | CC BY 4.0 | 300 test cases | v0.1.0 | None | |
| ArabicWER (arabicwer) | Tool | Apache-2.0 | 120 conformance fixtures in the dev split | v0.1.0 | None | |
| Contributor consent templates | Guide | CC BY 4.0 | — | v1.0 | None | |
| Dataset template and release kit | Tool | Apache-2.0 | — | v0.1.0 | None | |
| GCC Format Validators and Arabic ITN/TN Test Pairs | Dataset | CC BY 4.0 | 2000 ITN/TN test pairs | v0.1.0 | None | |
| Open-programme policy and operating manual | Report | CC BY 4.0 | 13 stages in the operating manual | v1.0.1 | None | |
| Arabic Agent Reliability Lab | Tool | Apache-2.0 | 21 synthetic workflow families | v0.2.0 | 10.5281/zenodo.23355782 | |
| Arabic Retrieval Starter Kit | Dataset | CC BY 4.0 | 12 passages in Arabic and English | v0.1 | 10.5281/zenodo.23355780 | |
| LabelBench | Benchmark | CC BY 4.0 | 12 model-task results | v0.1 | 10.5281/zenodo.23355784 |
Why we publish open Arabic data
Bayanat Labs is a Riyadh-based Arabic data company. We build data, never models, so we never compete with the teams that use our work. We publish small open releases, benchmarks, tools and guides so anyone can test Arabic AI work in the open, check our methods and cite the result.
Bayanat Labs is not affiliated with bayanat.ae (the UAE open-data portal), the former Bayanat AI (now Space42), or the ARBML bayanat library.
How the programme works
The rules behind every release: what we publish and under which licences, the stages a dataset passes before it ships, and the standard new datasets are built to.
Open-data policy →
Our release tiers, the licences we publish under and how we choose what to build.
How we build →
The scoping memo, the gated stages and the definition of done behind every open dataset.
Dataset standard →
The dataset template new releases are built from, and our checklist for Arabic dataset cards.
Programme pages
How to cite →
Concept and version DOIs, BibTeX and CITATION.cff for every release.
Claims register →
Every finding we state on these pages, with its date and scope.
Removal requests →
Ask us to remove or correct something. We answer at hello@bayanatlabs.com.
Our crawler →
How the BayanatLabs-OpenData crawler behaves and how to opt out.
For machines
Recent changes
| Arabic AI Model Tracker v0.1.0 | Register, exports, charts, Atom feed and latest.json on Hugging Face; crawler, inclusion rules and export builder on GitHub; the filterable release table on this site. | |
| Dataset Licence Compatibility Checker v0.1.0 | Rules engine, Python library, command line, GitHub Action, rules and verdict test cases on GitHub; the web app in English and Arabic on Hugging Face Spaces. | |
| Arabic Document Atlas v0.3.0 | Expanded frozen training release: 200,286-image union and 679 ACO source books with published images, with complete/partial/error coverage reported separately. Machine word annotations remain distinct from reference text; three exact-pixel duplicate occurrences supply lineage only. Reference-pair and community-text cohorts are unchanged. | |
| Arabic Document Atlas v0.2.0 | Expanded frozen training release with additional ACO page images, machine word annotations, explicit source-book coverage, and a separate duplicate-lineage configuration. Reference-pair and community-text cohorts are unchanged. | |
| Arabic Annotation Guidelines Template | Licence stated: CC BY 4.0. | |
| Arabic ASR Channel Pairs v0.1.0 | Clean and rendered audio, metadata, measurements and Whisper word error rate tables on Hugging Face; the renderer, conditions file and impulse responses on GitHub. | |
| Arabic Document Atlas v0.1.0 | Frozen training release: unique image union, ACO image/OCR records, separately licensed PDF/reference pairs, text-only Wikisource records and hash-bound source evidence. | |
| Arabic Text Normalizer (arabic-normalize) v0.1.0 | Pilot code (v0.1.0, not released yet) on GitHub, the in-browser tool on Hugging Face Spaces and the conformance fixtures on Hugging Face. |
Work on Arabic data with us
Native Arabic speakers, linguists and domain professionals can join the Talent Network for remote, paid-per-task Arabic data work. It is free to apply.
Need production-grade, rights-cleared data?
We publish synthetic fixtures, open benchmarks, results, code and guides for free. Data at scale is licensed or built to order: ready-to-license Arabic speech and text datasets with commercial AI-use rights written into the agreement, and custom collection, annotation and evaluation.