What Arabic-language AI models does Falcon offer?

Updated

Falcon offers five Arabic-focused models on falconllm.tii.ae: Falcon Arabic (general Modern Standard Arabic and dialect language model), Falcon-H1-Arabic (a high-performance Arabic model built on the H1 hybrid architecture), Falcon-Emirati (a 7-billion-parameter model tuned to Emirati vocabulary, expressions, and cultural context), Falcon-ASR (speech transcription for Emirati and Arabic audio), and Falcon-OCR-Arabic (text extraction from images). They are aimed at developers and organizations building Arabic-language applications, with Falcon-Emirati, Falcon-ASR, and Falcon-OCR-Arabic explicitly positioned around how the UAE "speaks, writes and reads."

The models at a glance

Model Modality What it is for Notable detail from the site
Falcon Arabic Text (LLM) Advanced language AI for Arabic speakers Built for Modern Standard Arabic and dialects
Falcon-H1-Arabic Text (LLM) High-performance Arabic AI for real-world use Described as "a complete departure from traditional large language models"
Falcon-Emirati Text (LLM) Emirati vocabulary, expressions, and cultural context 7-billion-parameter model, purpose-built for the UAE
Falcon-ASR Speech → text Emirati and Arabic speech transcription; multilingual voice services Positioned for the UAE's diverse communities
Falcon-OCR-Arabic Image → text Extracting Arabic text from images Lightweight model with leading text-extraction performance

How to choose between them

The decision mostly comes down to input type and language variant:

  • General Arabic text generation or understanding → start with Falcon Arabic if you need dialect coverage, or Falcon-H1-Arabic if you want the newer high-performance Arabic model. The site presents H1-Arabic as a step away from conventional LLM design, so it is the one to evaluate when performance on Arabic tasks is the priority.
  • UAE-specific content → Falcon-Emirati. Its 7B size makes it the compact option, and its tuning targets Emirati expressions and cultural context rather than Arabic in general.
  • Audio input → Falcon-ASR, for transcribing Emirati and Arabic speech and for voice services serving multilingual communities.
  • Images containing Arabic text → Falcon-OCR-Arabic, a lightweight extraction model.

A practical pattern is to combine them: for example, OCR a scanned Arabic document with Falcon-OCR-Arabic, pass the extracted text to Falcon Arabic or Falcon-H1-Arabic for understanding or summarization, and use Falcon-ASR where the source is a recording. Each model's page on the site ("Learn More" / "View More") is the place to confirm task fit before committing.

What the site does and does not tell you

The site describes each model's purpose, architecture notes (H1's hybrid Transformer–Mamba design, H1R 7B's attention-plus-state-space approach), and the Arabic/UAE focus of the newer releases. It does not publish pricing, licensing terms, or access instructions in the material available here, so treat availability and cost as things to verify on each model's page rather than assume. The site's stated framing—"making advanced AI accessible and available to everyone, everywhere"—describes intent, not a confirmed free tier.

Where these fit in the wider Falcon family

The Arabic models sit alongside Falcon's general-purpose releases: Falcon Perception (multimodal vision-and-language), Falcon H1R 7B (reasoning in math and coding), and Falcon-H1 (the hybrid architecture underlying the H1 line). If your task is not Arabic-specific, those general models are the relevant comparison; the Arabic line is the choice when language variant, script, or speech in Arabic is central to the use case.