SALAD-Bench: leaderboard
Hierarchical safety benchmark for LLMs, attacks, and defenses, with a three-level taxonomy and variants for standard, attack-enhanced, defense-enhanced, and multiple-choice safety evaluation.
Metric: Average Safety Score (%). Source: github.com. Source: huggingface.co. Saturation forecast: Estimated already saturated. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 2 | 93.91 | #761 |
| 2 | GPT-4 Turbo | 86.88 | #404 |
| 3 | Llama 2 13B Chat Base | 81.27 | #1422 |
| 4 | Llama 2 70B Chat (HF) | 81.22 | #1264 |
| 5 | GPT-3.5 Turbo | 81.01 | #865 |
| 6 | gemma-2B (IT) | 73.12 | #1559 |
| 7 | Qwen 1.5 4B Chat | 59.43 | #1467 |
| 8 | internlm2-chat-7B | 58.99 | #1121 |
| 9 | Llama 2 7B Chat | 57.38 | #1499 |
| 10 | Yi Chat 34B | 55.38 | #976 |
| 11 | Qwen 1.5 72B Chat | 54.88 | #776 |
| 12 | gemma-7B (IT) | 54.81 | #1496 |
| 13 | internlm2-chat-20B | 54.61 | #999 |
| 14 | Gemini 1.0 Pro | 54.17 | #933 |
| 15 | Yi 6B (Chat) | 52.76 | #1408 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=salad-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.