Arabic IFEval: leaderboard
Arabic instruction-following evaluation adapted from IFEval. The headline is the source-defined mean of Arabic prompt-level and instruction-level accuracy, rounded to one decimal place as a percentage.
Metric: Arabic Accuracy (%). Source: huggingface.co. Source: huggingface.co. Saturation forecast: Estimated already saturated. 40 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet | 75.9 | #350 |
| 2 | GPT-4o (2024-08-06) | 74.4 | #336 |
| 3 | GPT-4o Mini (2024-07-18) | 71.4 | #558 |
| 4 | Claude 3.5 Haiku | 70.9 | #568 |
| 5 | Qwen 2.5 72B Instruct | 70.7 | #454 |
| 6 | Gemma 2 27B (IT) | 63.6 | #645 |
| 7 | Qwen 2.5 32B Instruct | 63.6 | #498 |
| 8 | aya-expanse-32B | 61.4 | #692 |
| 9 | c4ai-command-r7B-12-2024 | 60.8 | #1161 |
| 10 | Llama 3.3 70B Instruct | 60.6 | #534 |
| 11 | Claude 3 Haiku | 59 | #803 |
| 12 | Gemma 2 9B (IT) | 56 | #788 |
| 13 | Llama 3.1 405B Instruct | 53.8 | #457 |
| 14 | aya-expanse-8B | 53.5 | #931 |
| 15 | Qwen 2.5 7B Instruct | 52.8 | #862 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=arabic-ifeval · How It Works · Data refreshed daily, snapshot 2026-10-20.