Jais 3 AI Models Launched for Arabic Language by MBZUAI and Inception42

Abu Dhabi:Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), through its Institute of Foundation Models (IFM), has partnered with Inception42 to introduce Jais 3, a series of AI models designed for the Arabic language. These models, part of the Jais family first unveiled in Abu Dhabi, are capable of understanding both spoken and written Arabic across various dialects. They can perform tasks such as transcribing conversations and creating documents and presentations.

According to Emirates News Agency, the Jais 3 models integrate six open language models with Jais Speech, a speech recognition model tailored for Arabic and its dialects. These models, ranging from 900 million to 375 billion parameters, are built on IFM's K2 Horizon, a fully open-source foundation model fleet released in September 2026. They are trained on Modern Standard Arabic and regional dialects, reflecting the linguistic and cultural nuances of over 400 million Arabic speakers.

Haonan Li, Senior Research Scientist at IFM, emphasized the importance of understanding Arabic in its everyday context, which can vary by geography and generation. Richard Morton, Vice President and Executive Director of IFM, highlighted the UAE's investment in open AI research and its impact on the Arabic-speaking world, aligning with the UAE National AI Strategy 2031.

The models were evaluated on 24 Arabic benchmarks, with the largest Jais 3 model leading in agentic tasks. Jais Speech achieved the lowest error rates, and the 7-billion- and 32-billion-parameter models received the highest Arabic scores among similar open models. Dr. Larry Murray from Inception42 noted that Jais 3 enhances practical business applications for Arabic AI, supporting complex workflows with diverse model sizes suitable for various computing budgets.

Jais Speech can transcribe Arabic and its dialects while recognizing language switches to English or French, maintaining accuracy in real-world conditions. It also identifies dialects and adds Arabic vowel marks to transcripts. The collaboration between IFM and Inception42 has advanced the capabilities of Arabic large language models, with Jais 3 leading across all model sizes. Ashish Koshy, CEO of Inception42, described Jais 3 as a significant advancement, developed through training on actual spoken and written dialects and cultural contexts.