The Technology Innovation Institute said in a statement that it has launched Falcon-Emirati, a 7-billion-parameter large language model purpose-built for the Emirati dialect including its unique vocabulary, expressions, idioms and cultural references. According to TII figures, the model attained an 84.83 percent score on the Alyah native Emirati Arabic benchmark that encompasses everyday language, figurative expressions, heritage knowledge and poetry.
Falcon-Emirati was developed by TII researchers building upon the Falcon-H1-Arabic foundation with targeted training on native Emirati content along with Modern Standard Arabic materials focused on Emirati culture and heritage. The institute supplemented this with synthetic data generated using Emirati-specific linguistic resources to overcome the challenge of limited written dialect material. This approach aligns with the institute’s earlier releases including Falcon Arabic in May 2025 and the Falcon-H1 Arabic family in January 2026.
Alongside the language model TII introduced Falcon-ASR, a compact 1.6-billion-parameter automatic speech recognition system that handles both Standard and Emirati Arabic in addition to English, French, Spanish and Portuguese. The institute also unveiled Falcon-OCR-Arabic, an efficient model engineered to pull Arabic text, tables, formulas and structured data from images and documents. These multimodal tools extend the Falcon series into practical applications for voice services and document processing across the UAE’s diverse population.
Publicly available benchmarks from prior TII launches show Falcon-H1 Arabic achieving leading positions on the Open Arabic LLM Leaderboard with its 7-billion-parameter version scoring 71.47 percent to surpass larger models from regional and global developers. TII data places the 34-billion-parameter variant at 75.36 percent exceeding even 70-billion-parameter competitors such as Meta’s Llama-3.3 and Alibaba’s Qwen. The institute’s work builds on a series of incremental improvements that have expanded from general Arabic support to highly specific dialect handling.
Dr. Najwa Aaraj, chief executive officer of TII, connected the new models to the organization’s sovereign AI strategy during the announcement at the Ai Everything Abu Dhabi event. The Technology Innovation Institute reported that the models aim to make advanced AI accessible to developers, researchers and everyday users while embedding local linguistic and cultural elements. The releases occurred on the first day of the Ai Everything Abu Dhabi conference.
The new Emirati-focused model will be accessible via the Falcon Chat platform for testing and integration purposes. TII has additionally published the Alyah benchmark dataset to encourage continued advancement in dialect-aware AI systems. Previous iterations of the Falcon family have included open components to promote wider innovation in the field.
