ThaiSafetyBench: leaderboard
Thai language and Thai cultural-context safety benchmark testing model resistance to harmful-content prompts across risk categories.
Metric: Overall ASR (imported). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Around November 2026. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 4.43 | #98 |
| 2 | Claude Sonnet 4.5 | 9.75 | #142 |
| 3 | Qwen 2.5 72B Instruct | 10.99 | #454 |
| 4 | Qwen 2.5 7B Instruct | 14.43 | #862 |
| 5 | GPT-4o | 16.04 | #343 |
| 6 | Llama 3.3 70B Instruct | 16.87 | #534 |
| 7 | Llama-SEA-LION-v3-8B-IT | 16.9 | #804 |
| 8 | Llama 3.2 3B | 26.08 | #1396 |
| 9 | Llama 3.2 1B | 37.66 | #1561 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=thaisafetybench · How It Works · Data refreshed daily, snapshot 2026-10-20.