Mistral Large 4 & Polars 2.0: Gateway & Streaming

Data Point

Falcon-Emirati-7B handles Gulf dialect beyond MSA

7B parameter model trained on native Emirati text, synthetic data constrained by dialect-specific glossaries, and cultural context—scores 84.83% on Alyah benchmark, outperforming larger multilingual models on dialectal nuance.

Developers building for Arabic-speaking markets now have a production-grade model that captures idioms, poetry references, and cultural context that MSA-only systems miss entirely. Eliminates post-processing or custom fine-tuning for Emirati dialect adaptation.

Replaces generic Arabic model fallback for Emirati use cases. Requires evaluation on Alyah benchmark if dialect accuracy matters; 7B size is practical for inference-cost-sensitive deployments. Worth trying now if your user base is UAE/Gulf-focused, but MSA models still cover broader Arabic markets cheaper.

• “A model that only knows MSA can translate every word of an Emirati sentence and still miss what it actually means”
• “Falcon-Emirati-7B scores 84.83% on Alyah, ahead of every other Arabic and multilingual model we compared it against, including several models many times its size”
• “Emirati is mostly a spoken dialect. It shows up far less in writing online than MSA, or even other Gulf and Levantine dialects”
• “We constrained it with strict rules and glossaries and dictionaries built specifically for Emirati vocabulary and grammar”