LOADING

Falcon‑Emirati

AI built for Emirati Arabic, culture and context

A 7-billion-parameter language model purpose-built for the vocabulary, expressions, idioms and cultural references that shape how Emirati communities communicate.

Try Falcon-Emirati on Falcon Chat
Overview

AI that understands the language of everyday life

Falcon-Emirati brings the Falcon family closer to the language people actually use in the UAE. Emirati Arabic carries its own vocabulary, expressions and cultural references, many of which cannot be understood through literal translation alone.

Built on Falcon-H1-Arabic, Falcon-Emirati was developed by TII researchers using native Emirati content, material covering Emirati culture and heritage, and Emirati-specific linguistic resources to better reflect how the dialect is naturally used.

Built for more than the words

01

Language & Dialect

Understands local vocabulary, idioms and everyday Emirati expressions.

02

Culture & Heritage

Built to better understand Emirati cultural references, heritage and context.

03

Natural Interaction

Designed to respond naturally in Emirati Arabic rather than defaulting to Modern Standard Arabic.

Benchmarking Emirati Intelligence

Leading performance across dialect, correctness and cultural understanding

Falcon-Emirati was evaluated across Alyah, open-ended generation and cultural-understanding tests. Across the models tested, Falcon-Emirati-7B led every reported metric.

84.83%Alyah accuracy1,173-item native Emirati Arabic multiple-choice benchmark
52.1%Dialect fidelityLLM-judged score for responding naturally in Emirati Arabic
85.57%Cultural understandingUAE subset of the ArabCulture-Dialogue evaluation

On Alyah, Falcon-Emirati scored 84.83%, ahead of ALLaM-7B-Instruct-preview (77.24%), Jais-2-8B-Chat (78.09%), Gemma-3-27B-it (74.68%) and Fanar-2-27B-Instruct (66.15%) in the comparison.

The strongest distinction is dialect fidelity: Falcon-Emirati scored 52.1%, compared with 5.3% for the next-best model shown, demonstrating a much greater ability to answer in Emirati Arabic rather than defaulting to Modern Standard Arabic.

Alyah accuracy (multiple-choice)
  • Falcon-Emirati-7B84.8%
  • ALLaM-7B-Instruct77.2%
  • Jais-2-8B-Chat78.1%
  • gemma-3-27b-it74.7%
  • Fanar-2-27B-Instruct66.2%
Dialect fidelity (LLM-judged, partial credit)
  • Falcon-Emirati-7B52.1%
  • ALLaM-7B-Instruct5.3%
  • Jais-2-8B-Chat2.0%
  • gemma-3-27b-it3.2%
  • Fanar-2-27B-Instruct0.4%
Head-to-head on the models tested across all three evaluations
ModelAlyah
(MCQ accuracy)
LLM-judged
correctness
LLM-judged
dialect fidelity
Cultural
understanding
Falcon-Emirati-7B84.83%48.8%52.1%85.57%
ALLaM-7B-Instruct-preview77.24%43.1%5.3%83.39%
Jais-2-8B-Chat78.09%38.3%2.0%73.79%
gemma-3-27b-it74.68%41.7%3.2%—
Fanar-2-27B-Instruct66.15%26.7%0.4%71.50%

Falcon-Emirati-7B leads on every metric above and by 10–130× on dialect fidelity, the only model that reliably answers back in Emirati instead of defaulting to MSA.

Cultural understanding by category
Radar charts comparing Falcon-Emirati-7B with Jais-2-8B-Chat, ALLaM-7B-Instruct-preview and Fanar-2-27B-Instruct across seven cultural categories

Falcon-Emirati-7B wins the majority of categories against all three competitors, by the widest margins on Poetry & Creative Expression and Language & Dialect and sweeps every category against Fanar-2-27B-Instruct. The one close category is Greetings & Daily Expressions, where Emirati and MSA overlap most.

Experience Falcon-Emirati

Emirati Arabic, understood in context.

Falcon-Emirati is available through Falcon Chat, enabling users to interact directly with a model designed around Emirati Arabic and its cultural context.