UAE Unveils State-of-the-Art Arabic AI Model
The United Arab Emirates has unveiled Jais 2, the largest and most advanced Arabic-centric artificial intelligence model developed to date. The launch marks a significant milestone in the UAE’s ongoing effort to strengthen its position as a global leader in AI research and innovation. Built through a collaboration between MBZUAI, Inception (a G42 company), and Cerebras Systems, the model aims to fill a longstanding gap in high-quality Arabic language AI tools.
Jais 2 is a 70-billion-parameter large language model, trained on the largest Arabic-first dataset ever assembled. The dataset includes modern standard Arabic, 17 regional Arabic dialects, and even Arabizi, the widely used Arabic written in Latin characters. With more than 600 billion Arabic tokens in its training mix, the model is designed to process and understand the language with exceptional depth and contextual accuracy.
Designed for Arabic and Beyond Built With an Arabic-First Approach
While global AI models often prioritize English, Jais 2 was built with an Arabic-first architecture to ensure accurate representation of linguistic diversity across the Arab world. Its training incorporates not only classical and modern standard Arabic but also dialects from the Gulf, Levant, North Africa, and beyond. This gives the model a strong foundation for real-world use cases in communication, education, media, digital services, and public-sector applications.
In addition to its Arabic proficiency, Jais 2 also includes a substantial English component in its training data, allowing it to perform bilingual tasks with high effectiveness. This makes it suitable for use in sectors where both languages are commonly used side by side.
Open-Weight Model to Accelerate Innovation
One of the standout features of Jais 2 is its open-weight availability, giving developers, researchers, and organisations free access to its model weights. This move is intended to accelerate AI adoption across Arabic-speaking markets by reducing dependence on Western, English-trained AI systems. Through open access, startups and enterprises can integrate the model into products, services, and enterprise solutions without licensing barriers.
Strong Benchmark Performance
Early testing shows that Jais 2 delivers state-of-the-art results in both Arabic and bilingual language tasks, outperforming previous Arabic-focused models. Its performance highlights the impact of training an LLM on a large-scale, purpose-built Arabic dataset instead of using general datasets that mix many languages.
Strengthening the Region’s AI Leadership
The launch of Jais 2 underscores the UAE’s commitment to advancing AI capabilities tailored to regional needs. By combining technical scale, linguistic depth, and open accessibility, Jais 2 is expected to support new innovation across industries and become a foundational tool for the next generation of Arabic digital transformation.
