SORRY-Bench - Hate Speech Generation: leaderboard
The 50 SORRY-Bench instructions that ask for insults, slurs, threats, obscenity or libel; the number is the share the model fulfils.
Metric: Fulfillment rate on unsafe instructions (%, lower is better). Source: sorry-bench.github.io. 55 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 2 9B (IT) | 0 | #2262 |
| 2 | Gemini 1.5 Pro | 0 | #1049 |
| 3 | Claude 3 Haiku (20240307) | 0 | #2314 |
| 4 | Claude 3 Sonnet (20240229) | 0 | #1911 |
| 5 | Llama 2 7B Chat | 0 | #8805 |
| 6 | Llama 2 70B Chat | 0 | #6800 |
| 7 | GPT-3.5 Turbo (1106) | 0 | #3214 |
| 8 | Llama 2 13B Chat | 0 | #8186 |
| 9 | Claude 2.1 | 0 | #2696 |
| 10 | Claude Instant 1.2 | 0 | #4019 |
| 11 | Claude 2 | 0 | #2427 |
| 12 | Gemini 1.5 Flash | 2 | #1668 |
| 13 | gemma-2B (IT) | 2 | #9327 |
| 14 | Claude 3.5 Sonnet (20240620) | 4 | #687 |
| 15 | Claude 3 Opus (20240229) | 4 | #1103 |
Interactive version: theaggregate.ai/benchmark?slug=sorry-bench-hate-speech-generation · How It Works · Data refreshed daily, snapshot 2026-11-13.