Back to Now
TII Falcon-Emirati Arabic model announcement launch Developing

TII announces Falcon-Emirati and new Arabic speech and document models

The Abu Dhabi research institute introduced models for Emirati Arabic conversation, speech transcription and Arabic document extraction; public availability varies.

Why now TII published its three-model announcement on October 6, with the official article indexed at 12:46 UTC.

Three models for different Arabic tasks

Abu Dhabi’s Technology Innovation Institute announced three new AI models on October 6. Falcon-Emirati is a seven-billion-parameter language model trained to handle the vocabulary, idioms and cultural context of Emirati Arabic. Falcon-ASR is a 1.6-billion-parameter speech-recognition model for Emirati and Modern Standard Arabic as well as English, French, Spanish and Portuguese. Falcon-OCR-Arabic is designed to extract Arabic text and structured material, including tables and formulas, from images and documents. These are separate language, speech and document tasks, rather than one model performing all three.

Availability and evidence

TII says Falcon-Emirati will be available through its Falcon Chat platform. Its announcement does not specify when all three models will be broadly accessible or provide a link to downloadable weights. TII reports that Falcon-Emirati scored 84.83% on its Alyah benchmark for native Emirati language and exceeded the open models it evaluated. That result is a company-reported comparison on a named test, not an independent ranking of every Arabic model. The useful next evidence will be public access, technical documentation and outside tests of how the systems handle dialects, noisy speech and complex documents.