Arabic Broad Leaderboard: leaderboard
Arabic Broad Benchmark: 470 human checked questions drawn from 64 Arabic datasets, covering 22 skills from grammar, dialect and diacritization to reasoning, translation and function calling. Each answer is scored 0 to 10 by a mix of manual rules and skill specific GPT-4.1 judging, and the board reports the average across questions, higher is better. Source: huggingface.co/blog/silma-ai/arabic-llm-leaderboard.
Metric: Average Score (0-10). Source: huggingface.co. Saturation forecast: Estimated already saturated. 121 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.5 Flash | 9.25 | #60 |
| 2 | Gemini 3.1 Pro (Preview) | 9.21 | #58 |
| 3 | Gemini 3 Pro (Preview) | 9.2 | #66 |
| 4 | Gemini 3.7 Flash | 9.2 | #23 |
| 5 | Muse Spark 1.1 | 9.2 | #52 |
| 6 | Gemini 3.1 Flash Lite | 9.14 | #191 |
| 7 | DeepSeek V4.1 Flash | 9.14 | #36 |
| 8 | Gemini 3.8 Flash | 9.14 | #21 |
| 9 | Gemini 4 Argon | 9.13 | #8 |
| 10 | GLM-5 | 9.06 | #141 |
| 11 | GPT-6.1 Pro Sol | 9.05 | #13 |
| 12 | GPT-4o | 9.02 | #343 |
| 13 | Gemini 2.5 Pro (Preview 05-06) | 8.99 | #147 |
| 14 | Gemma 4 31B (IT) | 8.98 | #176 |
| 15 | GPT-5.5 | 8.98 | #31 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=arabic-broad-leaderboard · How It Works · Data refreshed daily, snapshot 2026-10-20.