M-ALERT - Spanish: leaderboard
The model answers 15k red-teaming prompts in the Spanish translation across 32 risk categories; the score is the share LlamaGuard-3 labels safe. Comparing it with the English board shows where safety drops in translation.
Metric: Safe responses on the Spanish prompts (%, 15k prompts, LlamaGuard-3 judge, higher is better). Source: arxiv.org. 39 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 2 27B (IT) | 99.15 | #1782 |
| 2 | Llama 3.1 8B Instruct | 99.09 | #4246 |
| 3 | Gemma 2 9B (IT) | 99.06 | #2262 |
| 4 | Gemma 2 2B (IT) | 98.88 | #6876 |
| 5 | Qwen 2.5 14B Instruct | 98.87 | #1732 |
| 6 | Qwen 2.5 72B Instruct | 98.78 | #1092 |
| 7 | Qwen 2.5 32B Instruct | 98.7 | #1215 |
| 8 | Qwen 2.5 7B Instruct | 98.6 | #2564 |
| 9 | Qwen 2.5 3B Instruct | 98.56 | #5150 |
| 10 | GPT-4o (2024-11-20) | 98.28 | #866 |
| 11 | Llama 3.3 70B Instruct | 98.09 | #1357 |
| 12 | QwQ-32B | 97.85 | #910 |
| 13 | Gemini 2.0 Flash (001) | 97.68 | #727 |
| 14 | Mistral-Small-Instruct-2409 | 96.91 | #1991 |
| 15 | Qwen 2.5 1.5B Instruct | 96.41 | #7470 |
Interactive version: theaggregate.ai/benchmark?slug=m-alert-spanish · How It Works · Data refreshed daily, snapshot 2026-11-13.