The Technology Innovation Institute (TII), the applied research arm of Abu Dhabi’s Advanced Technology Research Council (ATRC), has launched three new artificial intelligence models designed to give developers, researchers and users access to advanced capabilities that reflect Emirati language and culture while supporting multimodal applications.
Leading the launch is Falcon Emirati, a seven-billion-parameter language model developed specifically to understand the Emirati dialect, including its vocabulary, expressions, idioms and cultural context.
TII has also introduced Falcon-ASR, a compact, 1.6-billion-parameter automatic speech recognition model that converts spoken language into written text. It supports Modern Standard Arabic, the Emirati dialect, English, French, Spanish and Portuguese.
The third model, Falcon Arabic OCR, is a lightweight, high-performance optical character recognition model designed to extract Arabic text and content from images and documents.
H.E. Faisal Al Bannai, Adviser to the UAE President for Strategic Research and Advanced Technology and Secretary-General of ATRC, said: “Our language is an integral part of the future we are building. With Falcon Emirati, we are translating this connection into technological capabilities that understand our society, reflect its culture and recognise its unique characteristics.”
He added: “Developing these capabilities here in the UAE gives us genuine freedom to continue advancing them, set our own priorities and turn our ambitions into practical solutions that enrich people’s lives and broaden their horizons.”
The new models are designed to enable smoother interaction with AI through text, speech and visual content, while strengthening the presence of Arabic within the wider AI ecosystem.
Falcon Emirati allows users to interact with AI in the Emirati dialect, while Falcon-ASR converts speech into text across languages widely used by UAE nationals, residents and international communities.
Potential applications include multilingual customer services, transcription of meetings and interviews, video subtitling, accessibility tools and voice-enabled digital services.
Dr Najwa Aaraj, Chief Executive Officer of TII, said developing sovereign AI capabilities requires models that understand language as people use it in everyday life. The Emirati dialect has its own expressions, cultural references and communication styles, she noted.
Through Falcon Emirati and Falcon-ASR, TII is working to equip the next generation of AI with a deeper understanding of Arabic and the way people in Emirati society communicate in practice, she added.
Falcon Emirati is built on the Falcon H1 Arabic model. In developing it, TII researchers used authentic Emirati content, Modern Standard Arabic material covering Emirati culture and heritage, and synthetic data derived from linguistic resources specifically designed for the Emirati dialect.
The model achieved a score of 84.83 per cent on Alyah, a benchmark developed specifically for the Emirati dialect. The benchmark assesses everyday language use, figurative expressions, and knowledge of heritage and poetry. According to TII, Falcon Emirati outperformed all open-source Arabic and multilingual models evaluated against the benchmark.
Falcon Emirati will be available to users through the Falcon Chat platform.
Falcon-ASR extends TII’s Arabic AI capabilities to spoken language, converting speech in the Emirati dialect and Modern Standard Arabic into written text, while also supporting English, French, Spanish and Portuguese.
Despite having only 1.6 billion parameters, the model recorded leading accuracy across general Arabic benchmarks and benchmarks focused on Emirati speech, outperforming larger models and systems evaluated under the same tests. In tests involving Emirati dialect speech, it also outperformed a multimodal model with 30 billion parameters.
Falcon-ASR can additionally identify the timestamp of each word in an audio file. This capability supports the creation of accurate video subtitles, searchable audio records, meeting and interview transcripts, accessibility tools and voice-enabled digital services.
The Falcon family is also expanding into Arabic text recognition and document understanding with Falcon Arabic OCR. The lightweight model is designed to extract text and structured data efficiently from images and documents, including tables, mathematical formulae and different document sections.
Its potential applications include document digitisation, converting archives into searchable content and automating information extraction.
Dr Hakim Hacid, Chief Researcher at TII’s Artificial Intelligence and Digital Science Research Centre, said the new capabilities would enable developers, researchers and communities across the region to build applications rooted in the Emirati dialect and culture, while opening up further opportunities for Arabic-language AI.
He said Falcon Emirati, Falcon-ASR and Falcon Arabic OCR had been developed to address specialised needs in language understanding, speech recognition and document processing — areas in which Arabic AI still has considerable room for growth and advancement.
The launch strengthens TII’s sovereign AI capabilities across language and multimodal technologies, reflecting the languages and communities these technologies are designed to serve.
