The Technology Innovation Institute (TII), Abu Dhabi’s applied research arm, has launched three Arabic AI models covering language, speech and document understanding.
The release includes Falcon-Emirati, a 7-billion-parameter language model trained for the Emirati Arabic dialect; Falcon-ASR, a speech transcription model; and Falcon-OCR-Arabic, which extracts Arabic text and structured information from images and documents.
Falcon-Emirati scored 84.83% on the Alyah benchmark, which tests everyday language, figurative expressions, heritage knowledge and poetry. According to TII, the result exceeded every Arabic and multilingual open-source model it evaluated. The model was trained on native Emirati content, material on Emirati culture and heritage, and synthetic data based on Emirati linguistic resources. It will be available through a Falcon Chat platform.
Falcon-ASR has 1.6 billion parameters and transcribes Standard and Emirati Arabic, English, French, Spanish and Portuguese. TII said it outperformed larger systems in its evaluation, including a 30-billion-parameter multimodal model on Emirati speech. It also adds word-level timestamps for subtitling, searchable recordings and transcription services.
Falcon-OCR-Arabic is designed to process tables, mathematical formulas and document sections, supporting document digitisation, searchable archives and automated information extraction.
Source: Middle East AI News
