List your company

Safety Evaluation on Jailbreaking Scenarios (Origin)

5.3Unsafe RatioDeepSeek-V3
4.5049.87715.2520.623Jan 22, 2025

Evaluation results

MethodLinks
DeepSeek-V3risk control system=true2025.015.325.4
DeepSeek-R1risk control system=true2025.018.527.3
o1release date=2024-12-172025.01950.4
Claude-3.7-Sonnet2025.0110.73.6
Qwen2.5 Instructparameters=72B2025.0113.85.4
DeepSeek-V3risk control system=false2025.0117.68.1
GPT-4orelease date=2024-05-132025.012217.1
DeepSeek-R1risk control system=false2025.0125.25.6