Datasets & benchmarks
Arabic datasets, benchmarks and leaderboards for training and testing models.
17 of 17 datasets
QCRI
ArabicCulturalQA
Cultural QA benchmark
MSAEgyptianLevantineGulfMaghrebiCross-dialect Arabic cultural question-answering benchmark (multiple-choice and open-ended), post-edited by native speakers. Non-commercial licence (CC BY-NC-SA 4.0).
Open · Free- Documented repository availability
- 10 May 2026
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
QCRI
MenaSpeechBank
Speech / conversational audio
MSAEgyptianMENA-centric speech bank: 17,641 utterances, 26 hours, 123 speakers across Arabic dialects, MSA and English. Non-commercial licence (CC BY-NC-SA 4.0).
Open · Free- Documented repository availability
- 3 Feb 2026
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
tiiuae
Alyah
Benchmark · 54 models tested
GulfMeasures how well models handle the Emirati dialect.
Open · Free- Release / announcement
- 27 Jan 2026
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
silma-ai
opensource-arabic-tts-benchmark
Benchmark
MSAOpen evaluation of Arabic text-to-speech models.
Open · Free- Documented repository availability
- 24 Jan 2026
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
SWivid
Habibi
Speech dataset
Regional dialectsHigh-quality Arabic audio for training voice models.
Open · Open- Documented repository availability
- 21 Jan 2026
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
stanford-crfm
HELM-Arabic
Leaderboard
MSATransparent, reproducible Arabic leaderboard built with Arabic.AI.
Open · Free- Release / announcement
- 18 Dec 2025
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
namaa-space
SAFIR
Benchmark
GulfSaudi Arabic task evaluation, part of the Saudi Dialect Hub.
Open · Free- Release / announcement
- 18 Nov 2025
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
ArabicNLP
AraHealthQA-2025
Benchmark
MSAFirst shared task on Arabic health question answering.
Open · Free- Release / announcement
- Nov 2025
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
ArabicNLP
AraGenEval
Benchmark
MSAArabic style transfer and AI-generated text detection.
Open · Free- Release / announcement
- Nov 2025
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
ArabicNLP
ImageEval-2025
Benchmark
MSAFirst Arabic image-captioning shared task.
Open · Free- Research paper published
- Nov 2025
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
QCRI
QASR
Speech recognition
MSAQCRI Al Jazeera speech corpus, about 2,000 hours of transcribed Arabic across dialects with code-switching. Non-commercial use only.
Open · Free- Documented repository availability
- 13 Oct 2025
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
QCRI
MultiNativQA
Multilingual QA
MSACulturally aligned native question-answering across 7 languages, including MSA and six Arabic dialects. Non-commercial licence (CC BY-NC-SA 4.0).
Open · Free- Documented repository availability
- 8 Oct 2024
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
QCRI
AraDiCE
Dialect & culture benchmark
MSAEgyptianLevantineBenchmarks for dialectal and cultural abilities of LLMs in Arabic, post-edited from standard benchmarks. Non-commercial licence (CC BY-NC-SA 4.0).
Open · Free- Documented repository availability
- 19 Sep 2024
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
OALL
Open-Arabic-LLM-Leaderboard
Leaderboard
MSACommunity ranking of open Arabic models by 2A2I, TII and Hugging Face.
Open · Free- Documented repository availability
- 14 May 2024
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
ClusterLab
arabic-101b-words
Text corpus · 101B words
MSACurated Arabic text dataset for training language models.
Open · Open- News published
- May 2024
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
QCRI
QADI
Dialect identification
DialectalQCRI Arabic Dialect Identification dataset: country-level Arabic dialects collected from Twitter.
Open · Free- Research paper published
- 19 Apr 2021
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
mozilla
common-voice-ar
Speech dataset
Regional dialectsCommunity-recorded Arabic speech; anyone can contribute.
Open · CC0- Release / announcement
- 12 Jul 2019
- Official source
- Official source — commonvoice.mozilla.org (opens in a new tab on commonvoice.mozilla.org)
- Item last updated
- Not documented yet
- Added to Souq
- Not recorded yet
- Verified pricing
- Not verified
- Commercial use
- Not verified
- Data residency
- Not verified
- Via Souq
- Discovery only
© 2026 Souq. Original text and design: all rights reserved. Third-party trademarks and source material belong to their respective owners.
This week in Arabic AI
New tools and the top news, once a week.