ThaiSafetyBench - Misinformation Harms: leaderboard
Metric: Attack Success Rate (%). Source: huggingface.co. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 7.84 |
| 2 | SeaLLMs-v3-7B-Chat | 10.63 |
| 3 | Qwen 2.5 72B Instruct | 12.54 |
| 4 | Claude Sonnet 4.5 | 13.41 |
| 5 | Qwen 2.5 7B Instruct | 17.25 |
| 6 | openthaigpt1.5-72B-instruct | 17.94 |
| 7 | Llama 3.3 70B Instruct | 18.12 |
| 8 | llama3.1-typhoon2-70B-instruct | 18.82 |
| 9 | Llama 3.2 3B Instruct | 21.43 |
| 10 | openthaigpt1.5-7B-instruct | 23.52 |
| 11 | Llama 3.1 8B Instruct | 27.18 |
| 12 | llama3.1-typhoon2-8B-instruct | 29.09 |
| 13 | llama3.2-typhoon2-3B-instruct | 29.97 |
| 14 | Llama 3.1 70B Instruct | 30.31 |
| 15 | Gemma 3 12B (IT) | 30.49 |
Interactive version: theaggregate.ai/benchmark?slug=thaisafetybench-misinformation-harms · How It Works · Data refreshed daily, snapshot 2026-09-19.