The Technology Innovation Institute (TII), the applied research arm of Abu Dhabi’s Advanced Technology Research Council (ATRC), has unveiled two major additions to its Falcon AI ecosystem: Falcon Arabic, the first Arabic-language model in the Falcon series, and Falcon-H1, a new hybrid-architecture model designed to redefine AI performance and portability. The announcement was made by H.E. Faisal Al Bannai, Advisor to the UAE President and Secretary General of ATRC, during a keynote at the Make it in the Emirates event in Abu Dhabi.

Built on the Falcon 3-7B architecture, Falcon Arabic is now the region’s best-performing Arabic language AI model. Trained on a high-quality, native (non-translated) Arabic dataset spanning Modern Standard Arabic and regional dialects, the model delivers state-of-the-art results across benchmarks such as the Open Arabic LLM Leaderboard, matching the performance of models up to 10 times its size. This development marks a significant milestone in the creation of sovereign, multilingual AI, enabling governments, researchers, and enterprises across the Arabic-speaking world to adopt AI solutions rooted in local language and cultural understanding. “We’re proud to finally bring Arabic to Falcon, and prouder still that the best-performing large language model in the Arab world was built in the UAE,” said H.E. Faisal Al Bannai.

In the 30B to 70B parameter category, Falcon-H1 outperforms Meta’s LLaMA and Alibaba’s Qwen on a range of benchmarks, including mathematics, reasoning, coding, long-context understanding, and multilingual tasks. But its standout innovation lies in accessibility and efficiency. Built with a hybrid architecture that combines the strengths of Transformers and Mamba, Falcon-H1 enables faster inference speeds, lower memory consumption, and real-world AI deployments on edge devices — a breakthrough for resource-limited settings. “Today, AI leadership is not about scale for the sake of scale,” said H.E. Faisal Al Bannai. “It is about making powerful tools useful, usable, and universal. Falcon-H1 reflects our commitment to delivering AI that works for everyone — not just the few.”

The Falcon-H1 model family includes variants ranging from 500M to 34B parameters, offering developers a flexible performance-to-efficiency spectrum: small models such as 500M and 1.5B are optimized for edge deployment, mid-range models like 3B and 7B balance capability with portability, and the flagship 34B model surpasses similarly sized LLaMA and Qwen models in complex tasks. “We approached Falcon-H1 not just as a research milestone but as an engineering challenge,” said Dr. Najwa Aaraj, CEO of TII. “How do we deliver exceptional efficiency without compromise? This model empowers innovators in environments where resources are limited but ambitions are not.” “This fundamentally shifts what’s possible at the smallest scale,” added Dr. Hakim Hacid, Chief Researcher at TII’s AI & Digital Science Research Center. “It enables powerful AI on edge devices, where privacy, low latency, and power efficiency are critical.”

Falcon models have already demonstrated global impact. In collaboration with the Bill & Melinda Gates Foundation, Falcon has powered AgriLLM, an AI solution that helps farmers make smarter agricultural decisions under challenging climate conditions. With over 55 million global downloads, the Falcon ecosystem is among the most widely adopted and highest-performing open-source AI frameworks to emerge from the Middle East. Unlike models built for narrow commercial use, Falcon models are foundational and versatile, designed for research, industry, education, and public-sector innovation. All models are open source, available via Hugging Face and FalconLLM.TII.ae under the TII Falcon License.

With Falcon Arabic and Falcon-H1, TII reinforces its commitment to building inclusive, sovereign, and impactful AI. By focusing on efficiency over scale, multilingual fluency, and real-world usability, these new models empower developers, governments, and innovators globally — especially in under-resourced regions. In an AI landscape increasingly defined by openness and utility, Falcon stands as a powerful force for democratized intelligence, rooted in excellence and engineered in the UAE.