Technology Innovation Institute (TII) has released Falcon-Emirati-7B, a 7-billion-parameter Arabic language model built on top of its Falcon-H1-Arabic family and adapted specifically for Emirati Gulf dialect, including its vocabulary, idioms, and cultural references rather than Modern Standard Arabic (MSA).
TII trained the model on authentic Emirati web and forum text, MSA material about Emirati culture and heritage, and synthetic dialectal data constrained by glossaries built for Emirati grammar and vocabulary.
On Alyah, a 1,173-question native benchmark for Emirati dialect capability, Falcon-Emirati-7B scores 84.83%, ahead of every other Arabic and multilingual model TII tested, including larger ones. A separate open-ended evaluation judged by Gemini 3.7 Flash measured whether models actually answer in Emirati dialect rather than defaulting to MSA: Falcon-Emirati-7B scored 0.52 on dialect fidelity, compared with 0.05 for ALLaM-7B-Instruct-preview, 0.03 for gemma-3-27b-it, 0.02 for Jais-2-8B-Chat, and effectively 0.00 for Fanar-2-27B-Instruct. The gap held across nearly every topic category, with the exception of basic greetings, where dialect and MSA overlap most.
On the UAE portion of the ArabCulture-Dialogue benchmark, which tests culturally appropriate responses, Falcon-Emirati-7B scored 85.57%, ahead of ALLaM-7B (83.39%), Jais-2-8B (73.79%), and Fanar-2-27B (71.50%).
TII notes the model can still reflect training-data biases and make mistakes on rare expressions or highly localized references, and recommends evaluation for specific use cases before relying on it for sensitive or official content. The model is available now on TII's chat platform.