Chat with Yehia
Navid AI's Arabic-first 7B model — was top-ranked on OALL, independently validated by KAUST and the Kingdom's BALSAM index, and built for everything from Modern Standard Arabic to dialect.
Benchmark Results
Rigorously evaluated across the most trusted Arabic AI benchmarks — OALL, AraLingBench (KAUST), and the Balsam Index.
OALL Average Score — 7B Model Class
Cross-Benchmark Comparison (AraLingBench)
Source: arxiv.org/abs/2511.14295| Model | AlGhafa | ArabicMMLU | EXAMS | MadinahQA | AraTrust | ALRAGE | ArbMMLU-HT | AraLingBench |
|---|---|---|---|---|---|---|---|---|
| Yehia-7B-preview ⭐ | 70.8 | 64.9 | 52.1 | 54.4 | 87.5 | 76.6 | 53.4 | 74.0 |
| ALLaM-7B-Instruct | 69.5 | 64.9 | 51.6 | 54.2 | 86.9 | 76.8 | 52.8 | 74.0 |
| Fanar-1-9B-Instruct | 76.4 | 65.8 | 52.7 | 73.4 | 88.3 | 77.0 | 58.6 | 60.0 |
| Qwen2.5-14B-Instruct | 72.3 | 60.0 | 53.6 | 35.6 | 86.1 | 78.9 | 55.7 | 58.7 |
| Hala-9B | 78.3 | 65.6 | 53.8 | 70.4 | 89.6 | — | 61.4 | 54.7 |
| Qwen2.5-7B-Instruct | 65.6 | 52.3 | 39.7 | 62.7 | 80.7 | 77.4 | 40.3 | 51.3 |
Recognized by the Kingdom's Official Arabic AI Benchmark
The BALSAM Index is developed by the King Salman Global Academy for Arabic Language (KSAA) in partnership with the Saudi Data & AI Authority — setting global benchmarks for evaluating AI models across Arabic NLP tasks including grammar, generation, comprehension, and more.
Yehia holds top positions across multiple BALSAM categories, demonstrating world-class Arabic language capabilities validated by the Kingdom's premier linguistic institution.
Paraphrasing
Ranked #1 globally across all models on text paraphrasing tasks
Subject Generation
2nd place globally on subject generation benchmarks
Sentiment Analysis
Top 3 globally on Arabic sentiment analysis tasks
Named Entity Recognition
Top 3 globally on NER tasks across Arabic text
Topic Identification
Top 3 globally on Arabic topic classification
Recognized by Leading Research
Yehia (Navid AI, 2025) has consistently ranked among top-performing Arabic models across multiple benchmarks in our evaluation.
AraLingBench Research Paper
King Abdullah University of Science & Technology
arxiv.org/abs/2511.14295 ↗The 3C3H Methodology
Yehia is trained with Group Relative Policy Optimization (GRPO), refined through six core principles that ensure every response meets the highest standards.
Correctness
Accurate information you can trust. Every fact is verified through rigorous evaluation loops.
Completeness
Full, well-rounded answers that cover the breadth and depth of every question.
Conciseness
Clear, to-the-point responses — no filler, no noise, just the information you need.
Helpfulness
Always aiming to support, uplift, and provide genuine value in every interaction.
Honesty
Transparent, straightforward communication — Yehia tells you what it knows and doesn't know.
Harmlessness
Promoting kindness and safety in every conversation, designed to be responsible by default.
Start Building with Yehia
Ready to power the next generation of Arabic AI applications.
TAKING AI FORWARD
Experience Products in AI
Empower your organization with Navid's AI-driven solutions, merging human expertise with advanced technology for unparalleled growth and security.
Copyright © 2024 Navid.