Jailbreak Evaluation on Multilingual Jailbreak Dataset (Evaluation set)
2.3JSRclaude-3-haiku
Evaluation Results
| Method | Links | |
|---|---|---|
| claude-3-haikuFamily=Claude, Safety Configuration=High_Risk2026.05 | 2.3 | |
| DeepSeekFamily=DeepSeek, Safety Configuration=Low-Creativity2026.05 | 2.9 | |
| DeepSeekFamily=DeepSeek, Safety Configuration=High-Risk2026.05 | 3 | |
| gemini-2.0-flashFamily=Gemini, Safety Configuration=Low_Creativity2026.05 | 3.5 | |
| gpt-4.1-miniFamily=GPT, Safety Configuration=Standard2026.05 | 4.2 | |
| claude-haiku-4.5Family=Claude, Safety Configuration=High_Risk2026.05 | 4.4 | |
| grok-4-fast-reasoningFamily=Grok, Safety Configuration=Low_Creativity2026.05 | 5.3 | |
| gpt-4o-miniFamily=GPT, Safety Configuration=Chaos2026.05 | 6.6 | |
| gemini-3-flash-previewFamily=Gemini, Safety Configuration=Chaos2026.05 | 11.8 | |
| grok-4-1-fast-non-reasoningFamily=Grok, Safety Configuration=Chaos2026.05 | 28.1 |