Belebele European - Italian: leaderboard
SpeakLeash Belebele European Leaderboard, Italian: Belebele multiple-choice reading comprehension over FLORES passages, four options per question, scored 0-shot by log-likelihood accuracy in an lm-evaluation-harness run. 0-100, higher is better, chance 25; open models of about 4.5B-15B parameters.
Metric: Accuracy (%, 0-shot). Source: huggingface.co. Saturation forecast: Estimated already saturated. 29 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | cogito-v1-preview-qwen-14B | 90.44 | |
| 2 | Qwen 3 14B | 90 | #542 |
| 3 | Gemma 2 9B (IT) | 89.67 | #795 |
| 4 | Qwen 2.5 14B Instruct | 88.44 | #654 |
| 5 | Gemma 3 12B (IT) | 87.22 | #672 |
| 6 | Qwen 3 8B | 86.22 | #678 |
| 7 | GaMS3-12B-Instruct | 85.89 | #706 |
| 8 | Qwen 2.5 7B | 83.22 | #895 |
| 9 | Phi-4 | 82.44 | #707 |
| 10 | Mistral Nemo Instruct (2407) | 82 | #917 |
| 11 | EuroLLM-9B-Instruct | 74 | #1047 |
| 12 | Apertus-8B-Instruct-2509 | 73 | #1028 |
| 13 | Gemma 2 9B | 71.44 | #924 |
| 14 | PLLuM-12B-nc-chat-250715 | 68.56 | #953 |
| 15 | EuroLLM-9B | 60.67 | #1238 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=belebele-european-italian · How It Works · Data refreshed daily, snapshot 2026-11-05.