LLM Trustworthy - Adversarial: leaderboard
The "Adversarial" subtask of LLM Trustworthy. AI-Secure's LLM trustworthiness leaderboard (DecodingTrust) — safety and reliability measured across adversarial robustness, ethics, fairness, out-of-distribution robustness, privacy, stereotype and toxicity dimensions, each reported as...
Metric: Trust Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3 8B Instruct | 67.28 | #1129 |
| 2 | Gemini 1.0 Pro | 67.28 | #933 |
| 3 | GPT-4 (0314) | 64.04 | #574 |
| 4 | Claude 2 | 57.98 | #761 |
| 5 | vicuna-7B-v1.3 | 52.16 | #1507 |
| 6 | GPT-4o (2024-05-13) | 51.36 | #325 |
| 7 | Llama 2 7B Chat | 51.01 | #1499 |
| 8 | GPT-4o Mini (2024-07-18) | 50.25 | #558 |
| 9 | Mistral-7B-OpenOrca | 47.23 | #1389 |
| 10 | mpt-7B-chat | 46.2 | #1556 |
| 11 | RedPajama-INCITE-7B-Instruct | 44.81 | #1566 |
| 12 | tulu-2-7B | 44.62 | #1429 |
| 13 | falcon-7B Instruct | 43.98 | #1622 |
| 14 | gemma-7B (IT) | 43.43 | #1496 |
| 15 | gemma-2B (IT) | 43.21 | #1559 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-trustworthy-adversarial · How It Works · Data refreshed daily, snapshot 2026-10-20.