ThaiSafetyBench - Thai Culture-Related Attacks: leaderboard
Metric: Attack Success Rate (%). Source: huggingface.co. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 5.28 |
| 2 | SeaLLMs-v3-7B-Chat | 10.64 |
| 3 | Claude Sonnet 4.5 | 11.19 |
| 4 | Qwen 2.5 72B Instruct | 13.99 |
| 5 | openthaigpt1.5-72B-instruct | 16.76 |
| 6 | Qwen 2.5 7B Instruct | 17.88 |
| 7 | GPT-4o | 20.89 |
| 8 | llama3.1-typhoon2-70B-instruct | 21.43 |
| 9 | openthaigpt1.5-7B-instruct | 21.48 |
| 10 | Llama 3.3 70B Instruct | 22.32 |
| 11 | Gemma 3 12B (IT) | 24.83 |
| 12 | Llama 3.2 3B Instruct | 29.96 |
| 13 | Llama 3.1 70B Instruct | 30.84 |
| 14 | Gemma 3 4B (IT) | 32.36 |
| 15 | llama3.2-typhoon2-3B-instruct | 35.37 |
Interactive version: theaggregate.ai/benchmark?slug=thaisafetybench-thai-culture-related-attacks · How It Works · Data refreshed daily, snapshot 2026-09-19.