MBZUAI and Inception42 Launch Jais 3 with Advanced Arabic AI Capabilities
The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) and Inception42 have launched Jais 3, a new generation of Arabic artificial intelligence models.
Developed through MBZUAI’s Institute of Foundation Models (IFM), Jais 3 focuses on Arabic language understanding, speech recognition, and AI-powered automation.
Moreover, the new models support regional dialects, multilingual conversations, and complex instructions, expanding AI accessibility across Arabic-speaking communities.
The Jais 3 family includes six open language models ranging from 900 million to 375 billion parameters. Additionally, it introduces Jais Speech, a dedicated Arabic speech recognition system.
Together, these technologies support applications such as conversation transcription, document creation, presentation development, and automated business workflows.
Jais 3 Expands Arabic AI Understanding Across Regional Dialects
Jais 3 builds on K2 Horizon, the fully open-source foundation model family that IFM released in September 2026.
The models understand Modern Standard Arabic alongside regional dialects, including Emirati, Gulf, and North African Arabic.
Furthermore, their development incorporates linguistic variations and cultural context to improve communication across diverse Arabic-speaking populations.
With more than 400 million Arabic speakers worldwide, the technology aims to address limitations in existing language models.
Additionally, Jais 3 supports conversations where speakers naturally switch between Arabic, English, and French.
This multilingual capability reflects everyday communication patterns across the Middle East and North Africa.
Meanwhile, researchers evaluated the models using 24 audited Arabic benchmarks, which they are releasing publicly alongside the technology.
According to the evaluation results, the largest Jais 3 model leads open models in agent-based tasks. It also matches leading models in Arabic language understanding.
Furthermore, the 7-billion- and 32-billion-parameter versions achieved the highest average Arabic scores among comparable open models.
These results highlight improvements in Arabic reasoning, instruction-following, and language processing across different model sizes.
Jais Speech Introduces Advanced Arabic Transcription and Recognition
Jais 3 also introduces Jais Speech, an Arabic speech recognition model that supports Modern Standard Arabic and dialects across 19 Arab countries.
Moreover, the system identifies regional dialects and transcribes conversations across different speaking conditions.
During evaluations, Jais Speech achieved the lowest average word and character error rates among the models tested.
Additionally, the technology preserves English and French words in their original scripts during multilingual conversations.
It can also add Arabic vowel marks to transcripts, improving readability and linguistic accuracy.
When combined with Jais 3 language models, the speech system enables users to issue spoken instructions in Arabic or English.
Consequently, AI applications can interpret requests and perform tasks through integrated language processing and automation capabilities.
Furthermore, the models maintain consistent language, tone, and communication styles across different Arabic dialects and registers.
These capabilities could support customer service, enterprise communications, transcription services, and multilingual digital assistants.
Jais 3 Strengthens UAE AI Innovation and Enterprise Automation
The Jais 3 launch represents another development in the UAE’s expanding artificial intelligence ecosystem.
Moreover, the initiative supports the UAE National AI Strategy 2031 by advancing local research capabilities and AI innovation.
The technology combines language understanding, reasoning, instruction-following, and tool usage to support increasingly sophisticated applications.
Consequently, businesses can develop AI-powered workflows that manage multiple tasks with reduced manual intervention.
Additionally, the availability of different model sizes allows organisations to select solutions according to their computing resources and deployment requirements.
Smaller models could support applications with limited infrastructure, while larger versions offer more advanced processing capabilities.
Furthermore, the open-model approach provides developers and researchers with opportunities to build specialised Arabic AI applications.
The focus on regional dialects and cultural context also addresses linguistic challenges that conventional multilingual AI systems may overlook.
Meanwhile, speech recognition and AI agent capabilities could accelerate digital transformation across industries requiring Arabic-language automation.
With Jais 3, MBZUAI and Inception42 aim to expand access to advanced Arabic AI technologies while strengthening the region’s research ecosystem.
Ultimately, the launch highlights the UAE’s growing role in developing language-focused artificial intelligence for businesses, developers, and Arabic-speaking communities worldwide.
