SALAD-Bench Attack: leaderboard
Metric: Safety Score (%). Source: huggingface.co. 34 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude 2 | 88.06 |
| 2 | GPT-4 Turbo | 80.28 |
| 3 | GPT-3.5 Turbo | 73.4 |
| 4 | Llama 2 70B Chat (HF) | 66.24 |
| 5 | Llama 2 13B Chat Base | 65.72 |
| 6 | gemma-2B (IT) | 50.34 |
| 7 | Yi 34B (Chat) | 23.64 |
| 8 | Qwen 1.5 4B Chat | 23.34 |
| 9 | Yi 6B (Chat) | 22.56 |
| 10 | internlm2-chat-7B | 20.28 |
| 11 | Gemini 1.0 Pro | 20.04 |
| 12 | Llama 2 7B Chat | 18.24 |
| 13 | Qwen 1.5 72B Chat | 16.22 |
| 14 | gemma-7B (IT) | 15.54 |
| 15 | chatglm3-6B | 12.48 |
Interactive version: theaggregate.ai/benchmark?slug=salad-bench-attack · How It Works · Data refreshed daily, snapshot 2026-09-05.