---
id: "2026-10-06-tii-falcon-emirati-falcon-asr-falcon-ocr-arabic"
url: "https://postcutoff.com/e/2026-10-06-tii-falcon-emirati-falcon-asr-falcon-ocr-arabic/"
as_of: "2026-10-12T22:30:00+02:00"
date: "2026-10-06"
date_precision: day
category: model-release
importance: 2
confidence: medium
status: [Partly confirmed]
sources: 4
editor: Adam Bicz
human_review: null
version: null
---

As of: 2026-10-12 22:30 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code). Human editor: Adam Bicz.
Canonical page: https://postcutoff.com/e/2026-10-06-tii-falcon-emirati-falcon-asr-falcon-ocr-arabic/

# TII launches Falcon-Emirati, Falcon-ASR and Falcon-OCR-Arabic

Full title: TII launches Falcon-Emirati (7B), Falcon-ASR (1.6B speech recognition) and Falcon-OCR-Arabic

Abu Dhabi's Technology Innovation Institute (TII) announced three Arabic-focused models on October 6, 2026: Falcon-Emirati, a 7B-parameter language model for Emirati Arabic; Falcon-ASR, a 1.6B-parameter speech-recognition model for Arabic, English, French, Spanish and Portuguese; and Falcon-OCR-Arabic for documents. TII reports a 20.92% average word error rate on the Open Universal Arabic ASR Leaderboard test sets. All scores are TII's own; no license or weights repository is stated.

## Key facts
- Falcon-Emirati: 7B parameters, built on Falcon-H1-Arabic; 84.83% on Alyah, an Emirati Arabic benchmark; available in Falcon Chat (chat.falconllm.tii.ae)
- Falcon-ASR: 1.6B parameters, builds on Falcon3-Audio (arXiv 2509.07526); word-level timestamps; one set of weights for all languages
- Falcon-ASR, Arabic: average WER 20.92% on six Open Universal Arabic ASR Leaderboard test sets, against a best published 23.17% (Hugging Face blog by TII)
- Falcon-ASR, internal Emirati test: WER 22.73%, CER 10.19%, 4.07 WER points better than Qwen3-Omni, the next best system; TII says it beat a 30B-parameter multimodal model
- Falcon-ASR, English: mean WER 5.74% on seven Open ASR Leaderboard test sets
- Access: a public demo Space (tiiuae/falcon-asr-demo); API access and native apps are planned; license and model repository not stated

## What happened

TII, the applied-research arm of Abu Dhabi's Advanced Technology Research Council, presented the three models as sovereign
language technology. TII chief executive Najwa Aaraj said: "Sovereign AI capability must reflect the language used in daily
life." Hakim Hacid, chief researcher of TII's AI and Digital Research Center, said the team "applied targeted specialization to
language, speech and document understanding". Falcon-OCR-Arabic extracts Arabic text, tables and formulas from images and
documents; TII gave no figures for it.

## Why it matters

Dialectal Arabic has far less transcribed speech than Modern Standard Arabic, so small models tuned for one dialect can beat
much larger general models on it. The comparison rests on TII's internal Emirati test set, which is not public.

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 159 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 98 days after its cutoff
- Gemini 3.8 Flash (training cutoff March 2026): 189 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 128 days after its cutoff

## Sources

1. [TII: TII launches Falcon-Emirati alongside new Arabic AI models for speech and visual text understanding](https://www.tii.ae/insights/tii-launches-falcon-emirati-alongside-new-arabic-ai-models-speech-and-visual-text) (tii.ae, official)
2. [Hugging Face blog (TII): Falcon-ASR](https://huggingface.co/blog/tiiuae/falcon-asr) (huggingface.co, official)
3. [Falcon-ASR demo Space](https://huggingface.co/spaces/tiiuae/falcon-asr-demo) (huggingface.co, code)
4. [RuntimeWire: TII says its 1.6B Falcon-ASR beats larger models on its Emirati test](https://runtimewire.com/article/tii-falcon-asr-emirati-arabic-speech-recognition) (runtimewire.com, press)

## Changes

- 2026-10-12 (filed): Created (evening)