QIMMA - ArabCulture: leaderboard

Metric: Accuracy (%, log-likelihood multiple choice). Source: huggingface.co. 83 models tracked.

Top models

#ModelScore
1Jais-2-70B-Chat83.24
2AceGPT-v2-32B-Chat79.79
3Llama 3.3 70B Instruct78.05
4jais-adapted-70B-chat78
5AIC-177.52
6Qwen 2.5 32B Instruct75.8
7Yehia-7B-preview75.47
8Hala-9B72.83
9jais-adapted-13B-chat72.16
10SILMA-9B-Instruct-v1.071.22
11Fanar-1-9B-Instruct70.71
12Jais-2-8B-Chat68.99
13LFM2 24B A2B68.83
14c4ai-command-r7B-arabic-02-202566.94
15Qwen 2.5 72B Instruct63.83

Interactive version: theaggregate.ai/benchmark?slug=qimma-arabculture · How It Works · Data refreshed daily, snapshot 2026-09-19.