Jais-3-7B: benchmark results
Provider: Inception (G42), Cerebras & MBZUAI. Released 2026-10-15. Access: Open.
Aggregate Rating 1655 ± 40, rank #588 of 10916 rated models, from 39 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Jais 3 Model Cards - Arabic IFEval | 80.2 | Accuracy (%) | 100 |
| Jais 3 Model Cards - Arabic IFBench (IFM) | 91.6 | Prompt-level strict accuracy (%) | 89.3 |
| Jais 3 Model Cards - BrowseComp-Plus | 67.7 | Accuracy (%) | 88.9 |
| Jais 3 Model Cards - QADI Dialect Identification | 69 | Accuracy (%) | 85.7 |
| Jais 3 Model Cards - QIAS-1 (IFM-audited) | 88.5 | Accuracy (%) | 82.1 |
| Jais 3 Model Cards - Arabic Linguistics - Diacritization (IFM) | 90.8 | Accuracy (%) | 78.6 |
| Jais 3 Model Cards - FannOrFlop MCQ | 69.4 | Accuracy (%) | 78.6 |
| Jais 3 Model Cards - SimpleQA Verified (Arabic) - Non-hallucination Rate | 68.3 | Non-hallucination rate (%) | 78.6 |
| Jais 3 Model Cards - FLORES-200+ Translation (LLM Judge) | 90.6 | LLM-as-a-judge score | 77.8 |
| Jais 3 Model Cards - Alyah (IFM-audited) | 80.7 | Accuracy (%) | 75 |
| Jais 3 Model Cards - Arabic Cuisine (IFM) | 85.5 | Accuracy (%) | 75 |
| Jais 3 Model Cards - IslamicEval 2025-1B (IFM-audited) | 74.9 | Accuracy (%) | 75 |
Interactive version: theaggregate.ai/model?slug=jais-3-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.