ThaiSafetyBench - General Prompt Attacks: leaderboard
Metric: Attack Success Rate (%). Source: huggingface.co. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 1.76 |
| 2 | Claude Sonnet 4.5 | 3.3 |
| 3 | openthaigpt1.5-72B-instruct | 3.67 |
| 4 | Qwen 2.5 72B Instruct | 3.83 |
| 5 | openthaigpt1.5-7B-instruct | 5.59 |
| 6 | SeaLLMs-v3-7B-Chat | 5.7 |
| 7 | Qwen 2.5 7B Instruct | 5.75 |
| 8 | GPT-4o | 5.96 |
| 9 | Llama 3.3 70B Instruct | 6.39 |
| 10 | Gemma 3 12B (IT) | 9.27 |
| 11 | llama3.1-typhoon2-70B-instruct | 10.06 |
| 12 | Llama 3.1 70B Instruct | 12.52 |
| 13 | Llama 3.1 8B Instruct | 15.34 |
| 14 | Gemma 3 4B (IT) | 17.65 |
| 15 | Llama 3.2 3B Instruct | 17.89 |
Interactive version: theaggregate.ai/benchmark?slug=thaisafetybench-general-prompt-attacks · How It Works · Data refreshed daily, snapshot 2026-09-19.