QIMMA - PALMX: leaderboard
PalmX multiple-choice questions on Arabic and Islamic culture, as run in the QIMMA Arabic leaderboard. Scored with LightEval log-likelihood over the answer options with character-length normalization rather than generated answers, so reasoning-tuned models can land near chance; accuracy 0-100, higher is better.
Metric: Accuracy (%, log-likelihood multiple choice). Source: huggingface.co. Saturation forecast: Around February 2027. 83 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 397B A17B | 83.91 | #154 |
| 2 | Jais-2-70B-Chat | 83.73 | #430 |
| 3 | Qwen 3.5 122B A10B | 81.46 | #184 |
| 4 | Gemma 4 31B | 81.23 | #212 |
| 5 | Qwen 3.5 35B A3B Base | 79.65 | #319 |
| 6 | Jais-2-8B-Chat | 78.98 | #802 |
| 7 | gemma-3-27B-pt | 78.62 | #505 |
| 8 | Llama 3.3 70B Instruct | 77.95 | #541 |
| 9 | Qwen 2.5 72B Instruct | 77.77 | #469 |
| 10 | ALLaM-7B-Instruct-preview | 76.84 | #724 |
| 11 | Gemma 4 26B A4B | 76.39 | #360 |
| 12 | AIC-1 | 76.11 | #235 |
| 13 | Qwen 3.5 27B | 75.93 | #192 |
| 14 | jais-adapted-70B-chat | 75.51 | #639 |
| 15 | AceGPT-v2-32B-Chat | 74.46 | #518 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=qimma-palmx · How It Works · Data refreshed daily, snapshot 2026-11-05.