arabic-nlp dialect-models emirati open-source benchmark

Falcon-Emirati-7B outperforms larger models on dialect

7B parameter model trained on authentic Emirati text and synthetic data achieves 84.83% on native dialect benchmark, replacing MSA-defaulting models for Gulf region applications.

Summary

Developers building for UAE/Gulf users now have a model that maintains dialect fidelity (0.52 dialect score vs 0.05 for competing 7B+ alternatives) instead of reverting to Modern Standard Arabic. Cuts integration friction for culturally-specific applications.

Why it matters

Implementation verdict

Replaces ALLaM-7B-Instruct and Jais-2-8B for Emirati use cases. Requires evaluation on your specific domain before production (TII explicitly warns on rare expressions and localized references). Worth trying now if targeting Gulf market; available on TII's chat platform today.

Sources

1.Falcon-Emirati-7B scores 84.83%, ahead of every other Arabic and multilingual model TII tested, including larger ones
2.Falcon-Emirati-7B scored 0.52 on dialect fidelity, compared with 0.05 for ALLaM-7B-Instruct-preview
3.built on top of its Falcon-H1-Arabic family and adapted specifically for Emirati Gulf dialect
4.trained on authentic Emirati web and forum text, MSA material about Emirati culture and heritage, and synthetic dialectal data
5.TII notes the model can still reflect training-data biases and make mistakes on rare expressions or highly localized references

Dev Signal

Get briefs like this in your inbox — free, every weekday.

100+ sources compressed into one 4-minute read. Ranked, cited, implementation-ready.

Read the full issue →All briefs