The Technology Innovation Institute (TII), the applied research arm of Abu Dhabi’s Advanced Technology Research Council (ATRC), has announced Falcon-H1 Arabic, a newly developed large language model built on a hybrid Mamba-Transformer architecture.
Representing a complete departure from previous transformer-based versions, this new model establishes itself as the highest-performing system on the Open Arabic LLM Leaderboard (OALL).
This milestone sets Falcon-H1 Arabic as the leading Arabic AI model currently available, outperforming models several times larger while delivering state-of-the-art accuracy, context handling, and linguistic representation.
Faisal al Bannai, adviser to the UAE president and secretary general of the Advanced Technology Research Council said,
“Falcon-H1 Arabic reflects our ongoing commitment to strengthening the UAE’s position as a global hub for advanced technology and responsible AI. By delivering models that support the linguistic and cultural needs of the region, we enable innovation that is accessible, relevant, and impactful across our societies. This achievement is a testament to the depth of talent and research expertise within TII.”
Building on the strong reception of the Falcon-Arabic models released earlier this year, which highlighted the community’s clear need for high-quality Arabic LLMs, TII has advanced its work with the new Falcon-H1 Arabic family.
Available in 3B, 7B, and 34B parameter sizes, the models are designed to meet diverse infrastructure and use-case needs.
Falcon-H1 Arabic introduces improvements in data quality, dialect coverage, long-context stability, and mathematical reasoning, enabling more accurate, reliable, and contextually aware Arabic understanding across real-world applications.
Dr. Najwa Aaraj, CEO of TII said,
“The development of Falcon-H1 Arabic builds on years of foundational work in Arabic AI and responds directly to the needs of our communities, including developer and businesses. By advancing architecture, data quality, and long-context reasoning, we are creating enablers that unlock new possibilities in education, healthcare, governance, and enterprise, and more, all in Arabic. This model represents an important step in our mission to deliver world-class AI that serves the region and contributes to global progress.”
Benchmark Results
On the OALL leaderboard – which evaluates models across a wide range of Arabic understanding and reasoning tasks – Falcon-H1 Arabic demonstrates clear performance leadership:
- The 3B model scores an average of 61.87%, 10 points ahead of leading 4B competitors, such as Microsoft’s Phi-4 Mini.
- The 7B model scores an average of 71.47%, surpassing all ~10B models, including Qatar’s Fanar-1-9B and Saudi Arabia’s HUMAIN ALLaM 7B model.
- The 34B model scores 75.36%, outperforming even 70B+ parameter systems, including China’s Qwen2.5 72B and META’s Llama-3.3 70B.
Beyond OALL, the Falcon-H1 Arabic models also achieve outstanding results on more targeted benchmarks, including,
- (i) 3LM, for STEM reasoning,
- (ii) ArabCulture, for cultural and contextual understanding, and
- (iii) AraDice (dialect comprehension).
Together, these results represent a breakthrough moment for Arabic AI. Falcon-H1 Arabic is not only outperforming models several times larger, across both general and specialized benchmarks, but is also demonstrating a level of linguistic depth, reasoning capability, and efficiency that sets a new benchmark for the field.
This establishes Falcon-H1 Arabic as the most capable and versatile Arabic language models developed to date.
Dr Hakim Hacid, Chief Researcher for TII’s Artificial Intelligence and Digital Research Centre (AIDRC) said,
“This model reflects our focus on building Arabic AI that is not only more advanced, but genuinely useful in real-world settings. By improving efficiency, depth of understanding and language coverage, we are enabling AI systems that can better support institutions, developers, and communities across the region.”
The model also extends context length dramatically, with windows of up to 256K tokens, enabling the models to work across large volumes of information in a single interaction.
In practice, this means users can, for example, analyze lengthy legal documents, medical notes, academic papers or enterprise knowledge bases without losing context or continuity – a capability that was not previously possible at this scale.
TII’s Falcon AI models have ranked number one across regional and global benchmarks since 2023, with Falcon-H1 Arabic now leading the Open Arabic LLM Leaderboard across model sizes.
These results demonstrate TII’s ability to build sovereign AI capabilities that compete at the highest global levels, while advancing Abu Dhabi and the wider UAE’s leadership in Arabic AI research and innovation.
The new models are now accessible within the playground through TII’s public interface.
