ThaiSafetyBench - Information Hazards: leaderboard
Metric: Attack Success Rate (%). Source: huggingface.co. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | SeaLLMs-v3-7B-Chat | 0.22 |
| 2 | GPT-5 | 0.55 |
| 3 | Claude Sonnet 4.5 | 0.66 |
| 4 | Qwen 2.5 72B Instruct | 1.1 |
| 5 | llama3.1-typhoon2-70B-instruct | 1.21 |
| 6 | openthaigpt1.5-72B-instruct | 1.32 |
| 7 | Qwen 2.5 7B Instruct | 1.87 |
| 8 | openthaigpt1.5-7B-instruct | 2.2 |
| 9 | Llama 3.3 70B Instruct | 3.08 |
| 10 | GPT-4o | 3.52 |
| 11 | Llama 3.1 70B Instruct | 3.96 |
| 12 | Llama 3.2 3B Instruct | 4.73 |
| 13 | Llama 3.1 8B Instruct | 4.84 |
| 14 | Gemma 3 12B (IT) | 9.45 |
| 15 | llama3.2-typhoon2-3B-instruct | 11.21 |
Interactive version: theaggregate.ai/benchmark?slug=thaisafetybench-information-hazards · How It Works · Data refreshed daily, snapshot 2026-09-19.