Falcon-H1 Arabic Sets New Global Benchmark in Arabic AI

Abu Dhabi, United Arab Emirates – 5 January 2026:

The Technology Innovation Institute (TII), the applied research arm of Abu Dhabi’s Advanced Technology Research Council (ATRC), has announced the launch of Falcon-H1 Arabic, a next-generation large language model that sets a new global standard for Arabic artificial intelligence.

Built on an innovative hybrid Mamba-Transformer architecture, Falcon-H1 Arabic represents a major evolution from earlier transformer-only models. This new design enables significantly higher accuracy, stronger reasoning, and superior long-context understanding, while operating with far fewer parameters. As a result, Falcon-H1 Arabic delivers exceptional performance and efficiency, establishing itself as the highest-ranked model on the Open Arabic LLM Leaderboard (OALL).

This achievement positions Falcon-H1 Arabic as the most advanced Arabic AI model currently available, outperforming competing systems that are several times larger in size. The model demonstrates state-of-the-art capabilities in linguistic representation, contextual comprehension, and reasoning across a wide range of real-world use cases.

His Excellency Faisal Al Bannai, Adviser to the UAE President and Secretary General of ATRC, stated:
“Falcon-H1 Arabic reflects our long-term commitment to strengthening the UAE’s role as a global leader in advanced technology and responsible AI. By delivering world-class models that address the linguistic and cultural needs of our region, we are enabling innovation that is inclusive, relevant, and impactful. This milestone highlights the depth of talent and research excellence within TII.”

Following the strong adoption of the earlier Falcon-Arabic models launched this year, TII advanced its research to meet the growing demand for high-quality Arabic large language models. The Falcon-H1 Arabic family, available in 3B, 7B, and 34B parameter sizes, is designed to support a broad range of infrastructure environments and application needs.

The new models introduce major enhancements in training data quality, Arabic dialect coverage, long-context stability, and mathematical and logical reasoning. These improvements enable more accurate, reliable, and context-aware Arabic language understanding across sectors such as education, healthcare, government services, and enterprise applications.

Dr Najwa Aaraj, CEO of TII, commented:
“Falcon-H1 Arabic builds on years of foundational research in Arabic AI and directly addresses the needs of developers, institutions, and businesses. By advancing model architecture, reasoning depth, and long-context capabilities, we are unlocking new possibilities across critical sectors—entirely in Arabic. This marks an important step in our mission to deliver world-class AI that serves the region while contributing to global progress.”

Benchmark Leadership

On the Open Arabic LLM Leaderboard (OALL), which evaluates models across comprehensive Arabic language understanding and reasoning tasks, Falcon-H1 Arabic demonstrates clear leadership:

• The 3B model achieves an average score of 61.87%, outperforming leading 4B models by more than 10 points.

• The 7B model records 71.47%, surpassing all comparable ~10B models, including prominent regional competitors.

• The 34B model reaches 75.36%, exceeding even 70B+ parameter systems from leading global AI developers.

Beyond OALL, Falcon-H1 Arabic also excels in specialized benchmarks, including 3LM for STEM reasoning, ArabCulture for cultural and contextual understanding, and AraDice for Arabic dialect comprehension.

Together, these results mark a breakthrough moment for Arabic AI. Falcon-H1 Arabic not only delivers unmatched performance across general and domain-specific benchmarks, but also achieves this with unprecedented efficiency, setting a new reference point for the field.

Dr Hakim Hacid, Chief Researcher at TII’s Artificial Intelligence and Digital Research Centre (AIDRC), said:
“Our focus is on building Arabic AI that is not just more powerful, but genuinely useful in real-world contexts. By enhancing efficiency, depth of understanding, and language coverage, Falcon-H1 Arabic enables AI systems that can better support institutions, developers, and communities across the region.”

A defining feature of Falcon-H1 Arabic is its extended context window of up to 256,000 tokens, allowing users to analyze large volumes of information—such as legal documents, medical records, academic research, or enterprise knowledge bases—within a single interaction, without losing coherence or continuity.

Since 2023, TII’s Falcon AI models have consistently ranked at the top of regional and global benchmarks. With Falcon-H1 Arabic now leading the Open Arabic LLM Leaderboard across all model sizes, TII further reinforces Abu Dhabi and the UAE’s position as a global hub for sovereign, high-performance Arabic AI innovation.