Abu Dhabi’s Technology Innovation Institute (TII) has unveiled Falcon-H1 Arabic AI, a groundbreaking large language model that establishes itself as the world’s most advanced Arabic AI system. Built on a hybrid Mamba-Transformer architecture, this model tops the Open Arabic LLM Leaderboard while outperforming competitors several times larger.

Breakthrough Performance with Falcon-H1 Arabic AI Architecture

The Falcon-H1 Arabic AI represents a complete departure from previous transformer-based versions, introducing a hybrid approach that delivers higher accuracy with smaller parameter sizes. Available in 3B, 7B, and 34B parameter configurations, the model addresses diverse infrastructure and use-case requirements across the region.

His Excellency Faisal al Bannai, Adviser to the UAE President and Secretary General of the Advanced Technology Research Council, emphasized that this achievement reflects the UAE’s commitment to strengthening its position as a global technology hub while delivering AI that serves linguistic and cultural needs.

Record-Breaking Benchmark Results Establish Global Leadership

On the Open Arabic LLM Leaderboard, the Falcon-H1 Arabic AI demonstrates unprecedented performance across all model sizes:

  • The 3B model scores 61.87%, exceeding leading 4B competitors like Microsoft’s Phi-4 Mini by 10 points
  • The 7B model achieves 71.47%, surpassing all 10B models including Qatar’s Fanar-1-9B
  • The 34B model reaches 75.36%, outperforming 70B+ systems including China’s Qwen2.5 72B and META’s Llama-3.3 70B

Dr. Najwa Aaraj, CEO of TII, noted that the development builds on years of foundational Arabic AI research, responding directly to community needs while enabling new possibilities in education, healthcare, governance, and enterprise applications.

This model reflects our focus on building Arabic AI that is not only more advanced, but genuinely useful in real-world settings.

Dr Hakim Hacid, Chief Researcher for TII’s AIDRC

Extended Context and Real-World Applications

The model extends context length dramatically to 256K tokens, enabling analysis of lengthy legal documents, medical notes, academic papers, and enterprise knowledge bases without losing continuity. Users can now access the Falcon-H1 Arabic AI through TII’s public interface at https://chat.falconllm.tii.ae, marking a significant advancement in Arabic language AI capabilities.