What changed
TII launched Falcon-Emirati, a 7B Emirati Arabic model; Falcon-ASR, a 1.6B speech recognizer for Standard and Emirati Arabic plus English, French, Spanish and Portuguese; and Falcon-OCR-Arabic for Arabic text, tables, formulas and document structure. Public demos and later OCR weights make access uneven across the three releases.
Why it matters
The release targets dialect, speech and document gaps that general multilingual systems often blur. Teams can now test localized interfaces and transcription or OCR paths, but should separate public demo access, downloadable artifacts and internal benchmark claims.
What to do
Use synthetic or consented Emirati speech and documents to test the public demos or released artifact. Record dialect coverage, WER or extraction error by document type, timestamp quality, privacy terms and whether the exact model is downloadable before selecting a production path.
Caveat
TII's headline accuracy is vendor-reported; part of the Emirati ASR evaluation is internal. Falcon-ASR's page said API and native applications were planned, while access across Falcon-Emirati and OCR used different demo or artifact paths.