ThaiSafetyBench - Malicious Uses: leaderboard
Metric: Attack Success Rate (%). Source: huggingface.co. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 1.06 |
| 2 | Claude Sonnet 4.5 | 4.39 |
| 3 | SeaLLMs-v3-7B-Chat | 6.8 |
| 4 | Qwen 2.5 72B Instruct | 8.61 |
| 5 | openthaigpt1.5-72B-instruct | 9.21 |
| 6 | GPT-4o | 9.67 |
| 7 | Gemma 3 12B (IT) | 10.27 |
| 8 | Qwen 2.5 7B Instruct | 10.88 |
| 9 | openthaigpt1.5-7B-instruct | 11.78 |
| 10 | Llama 3.3 70B Instruct | 13.44 |
| 11 | Gemma 3 4B (IT) | 17.54 |
| 12 | llama3.1-typhoon2-70B-instruct | 22.05 |
| 13 | Llama 3.1 8B Instruct | 25.08 |
| 14 | Llama 3.1 70B Instruct | 25.57 |
| 15 | Llama 3.2 3B Instruct | 28.1 |
Interactive version: theaggregate.ai/benchmark?slug=thaisafetybench-malicious-uses · How It Works · Data refreshed daily, snapshot 2026-09-19.