ThaiSafetyBench: leaderboard

Thai language and Thai cultural-context safety benchmark testing model resistance to harmful-content prompts across risk categories.

Metric: Overall ASR (self-reported). Source: benchmarklist.com. 15 models tracked.

Top models

#ModelScore
1GPT-54.43
2Claude Sonnet 4.59.75
3Qwen 2.5 72B Instruct10.99
4Qwen 2.5 7B Instruct14.43
5GPT-4o16.04
6Llama 3.3 70B Instruct16.87
7Llama 3.2 3B26.08
8Llama 3.2 1B37.66

Interactive version: theaggregate.ai/benchmark?slug=thaisafetybench · How It Works · Data refreshed daily, snapshot 2026-09-05.