Datasets & benchmarks

AI that tracks topics and trends in Arabic online conversations across the Arab digital sphere.

Tests how well 50+ models handle the Emirati dialect.

Open evaluation of Arabic text-to-speech models.

High-quality Arabic speech dataset for training voice models.

Transparent, reproducible Arabic LLM leaderboard, built with Arabic.AI.

Saudi Arabic task evaluation, part of the Saudi Dialect Hub.

First shared task on Arabic health question answering.

Shared task on Arabic style transfer and detecting AI-written Arabic text.

First Arabic image-captioning shared task.

An Arabic speech dataset pairs audio samples with transcriptions for speech research.

An Arabic embedding benchmark evaluates models across multiple tasks, domains and dialects.

Community ranking of open Arabic models, from 2A2I, TII and Hugging Face.

101-billion-word curated Arabic text dataset for model training.

An Arabic evaluation dataset probes language-model trustworthiness across several risk categories.

An Arabic multiple-choice benchmark tests language-model knowledge across academic subjects.

Community-recorded Arabic speech you can contribute to.